Catalog updates

Reverse-chronological log of changes to the catalog. Newest at the top.

2026-06-20

Chemistry directed pass plus a manifest sweep. The official MCP Registry surfaced CovaSyn Chemistry MCP (com.covasyn/chemistry, v1.27.2, registered 2026-06-02) — a hosted streamable-HTTP server with 130+ deterministic tools spanning ADMET, docking, retrosynthesis, ICH M7 toxicology, NMR/MS, and biologics design. It is installable today (free tier + npx @covasyn/mcp-client stdio proxy / remote https://mcp.covasyn.com/mcp), so it was added as a new page; it closes the long-standing “no Claude-installable retrosynthesis/ADMET wrapper” gaps noted in the curator state. The RDKit (rdkit-mcp, rdkit-skill), ChEMBL, PubChem, ChemLint, and ChemCP seed-query hits are already catalogued; AiZynthFinder/ASKCOS still ship only REST APIs with no discrete MCP/Skill, and Polaris has no Claude wrapper. User request #42 (open-targets) could not be actioned this run — the .request-bodies/42.md prefetch was missing, so the entry stays open per protocol.

Added

CovaSyn Chemistry MCP (Categories: Chemistry, Drug Repurposing and Discovery) — hosted deterministic cheminformatics MCP (130+ tools: ADMET, docking, retrosynthesis, ICH M7, NMR/MS, biologics); freemium, npx @covasyn/mcp-client or remote https://mcp.covasyn.com/mcp (source).

Verified (no changes)

Chemistry seed queries (RDKit, retrosynthesis, ChEMBL, Polaris, cheminformatics MCP) re-run; all installable hits already catalogued.

2026-06-14

Drug Repurposing and Discovery directed pass plus a manifest sweep. The anthropics/life-sciences marketplace matched the catalogued set (PubMed, BioRender, Synapse, Scholar Gateway, Consensus, Cortellis, AdisInsight, 10x Genomics Cloud, the four skills, plus the noted medidata/tooluniverse — all have pages). The directed pass confirmed the official Open Targets MCP and the ToolUniverse MCP server are already catalogued, but found that the ToolUniverse repo also ships ~68 discrete agent skills under skills/ (installable via npx skills add mims-harvard/ToolUniverse) that are distinct from the MCP-server entry. Added the Drug Repurposing skill (tooluniverse-drug-repurposing) as a new page; deferred the sibling drug-discovery skills. DrugBank and ADMET seed queries surfaced only already-catalogued or non-installable targets.

Added

Drug Repurposing (Claude Skill) (Categories: Drug Repurposing and Discovery, Translational Medicine) — ToolUniverse agent skill identifying repurposing candidates via target-, compound-, and disease-driven strategies with mechanism and feasibility scoring; npx skills add mims-harvard/ToolUniverse (requires the ToolUniverse MCP server) (source).

Verified (no changes)

anthropics/life-sciences marketplace re-fetched; all 13 entries already catalogued. Official Open Targets MCP (open-targets.md) and ToolUniverse MCP server (tooluniverse.md) confirmed current.

2026-06-14

Translational Medicine directed pass plus a manifest sweep of anthropics/life-sciences. Re-fetching the marketplace showed the medidata plugin now publishes a concrete HTTP MCP install path (medidata/.claude-plugin/plugin.json → https://mcp.imedidata.com/mcp), so the long-deferred Medidata Connector was promoted to a full page. The remaining clinical entries in the marketplace (clinical-trials, clinical-trial-protocol, cortellis, adisinsight) already have pages. FHIR (fhir-developer, fhir-wso2), CMS coverage, OpenFDA, and ClinicalTrials.gov directed-pass targets all matched catalogued entries; no other new installable Translational Medicine tool surfaced this slot.

Added

Medidata Connector (Categories: Translational Medicine) — Medidata’s hosted clinical-trial MCP: Platform Help (Knowledge Hub Q&A for Rave EDC / Data Connect / Clinical Data Studio) and Predictive Site Ranking (rank trial sites by predicted enrollment during protocol planning); /plugin install medidata@life-sciences or claude mcp add --transport http medidata https://mcp.imedidata.com/mcp (source).

Verified (no changes)

anthropics/life-sciences marketplace re-fetched; aside from medidata all clinical/translational plugins already catalogued.

2026-06-14

Neuroscience directed pass plus a manifest sweep. The anthropics/life-sciences marketplace matched the catalogued set (15 plugins/skills/connectors already have pages; medidata and tooluniverse remain noted/deferred); anthropics/claude-plugins-official contains no life-science plugins. The Neuroscience pass surfaced SpikeLab (braingeneers/SpikeLab, MIT) — a suite of computational-neuroscience agent skills for MEA spike-train data including a Kilosort2/4/RT-Sort spike-sorter skill and an optional built-in MCP server — added as a new page. The Allen Brain Atlas (allenbrain-mcp, already flagged on license) and AIND data MCP queries matched catalogued entries. While re-verifying, found that Neurosift Tools MCP now ships a .claude-plugin marketplace, so its page gained a /plugin install neurosift-tools@neurosift-mcps path alongside the existing manual build. The mne-neurophysiology-analysis mcpmarket listing and dandi-query-mcp boilerplate remain without a verifiable real tool surface and stay deferred.

Added

SpikeLab (Categories: Neuroscience) — Braingeneers agent-skill suite for MEA spike-train analysis and spike sorting (Kilosort2/Kilosort4/RT-Sort); pip install "spikelab[all]" + copy agent/skills/ into ~/.claude/skills/, with an optional MCP server via the mcp extra (source).

Updated

Neurosift Tools MCP — added a Claude Code plugin-marketplace install path (/plugin marketplace add + /plugin install neurosift-tools@neurosift-mcps) and a Claude Desktop stdio JSON snippet; last_verified bumped (repo confirmed active per upstream README).
AIND Data MCP — re-verified (repo active, MIT, uvx/uv tool install aind-data-mcp); last_verified bumped, no field changes.

Verified (no changes)

1 additional Neuroscience entry spot-checked.

2026-06-13

Molecular and Cellular Biology directed pass plus a manifest sweep. The anthropics/life-sciences marketplace matched the catalogued set (the medidata plugin remains gated behind account onboarding; clinical-trial-protocol-skill is already a page); anthropics/claude-plugins-official contains no life-science plugins. The directed pass surfaced the scmcphub single-cell MCP ecosystem — the already-catalogued scanpy-mcp has three sibling servers not yet in the catalog: LIANA-MCP (cell-cell communication), decoupler-MCP (pathway / TF activity), and CellRank-MCP (cell-fate / trajectory). Each is a discrete pip-installable MCP following the same <pkg> run stdio pattern, so all three were added as their own entries. No Bioconductor MCP and no standalone CRISPR-design MCP/Skill with a verifiable primary source surfaced this pass.

Added

LIANA-MCP (Categories: Immunology and Microbiology, Molecular and Cellular Biology) — scmcphub MCP wrapping liana-py for cell-cell communication (multi-method ligand-receptor inference, rank aggregation, circle/dotplot); pip install liana-mcp → liana-mcp run (source).
decoupler-MCP (Categories: Immunology and Microbiology, Molecular and Cellular Biology) — scmcphub MCP wrapping decoupler for pathway (PROGENy) and transcription-factor (CollecTRI) activity inference; pip install decoupler-mcp → decoupler-mcp run (source).
CellRank-MCP (Categories: Molecular and Cellular Biology) — scmcphub MCP wrapping CellRank for cell-fate / trajectory modeling (kernels, GPCCA, terminal/initial states, fate probabilities); pip install cellrank-mcp → cellrank-mcp run (source).

2026-06-13

Integrative Structural and Computational Biology directed pass plus a manifest sweep. The anthropics/life-sciences marketplace matched the catalogued set (all 21 plugins/skills/connectors already have pages); anthropics/claude-plugins-official contains no life-science plugins. The Structural pass surfaced one genuinely-new structural-biology MCP — cyanheads/protein-mcp-server (Apache-2.0, in the official MCP Registry), which orchestrates RCSB PDB + PDBe + UniProt with structure-similarity and ligand-tracking tools — folded into the existing pdb.md page as a third install option (one entry per data source). The ChatMol GROMACS/PyMOL/ChimeraX MCP and the AlphaFold DB MCP that the seed queries surfaced are already catalogued. A Foldseek structure-search skill (mcpmarket) and cryo-EM (RELION/cryoSPARC/EMDB) wrappers remain without verifiable primary sources and were deferred.

Added

PDB MCP Server (Categories: Integrative Structural and Computational Biology, Drug Repurposing and Discovery) — added cyanheads/protein-mcp-server (Apache-2.0) as a third install option on the existing page: a multi-provider MCP orchestrating RCSB PDB + PDBe + UniProt with protein_search_structures / protein_get_structure / protein_find_similar / protein_track_ligands (source).

Verified (no changes)

AlphaFold MCP Server — re-verified upstream repo active, LICENSE present, ~20+ tools, install path unchanged (last_verified left at 2026-05-20 pending the next full sweep).

2026-06-13

Immunology and Microbiology directed pass plus a manifest sweep. The anthropics/life-sciences marketplace matched the catalogued set — every in-scope plugin/skill/connector is already a page. The Immunology pass surfaced one genuinely-new installable MCP server (Human Protein Atlas, an antibody-based resource). IEDB, BCR/TCR-repertoire (Immcantation), and metagenomics (Kraken2/QIIME2/MetaPhlAn) still lack discrete Claude-installable wrappers — BioinfoMCP is a converter framework, not a discrete tool — and were re-deferred; BoltzGen (binder/antibody design) was deferred as an unverifiable wrapper over a model-weight artifact.

Added

Human Protein Atlas MCP Server (Categories: Immunology and Microbiology, Molecular and Cellular Biology, Translational Medicine) — MIT; Augmented-Nature/ProteinAtlas-MCP-Server, 14 read-only tools over the Human Protein Atlas (protein search, tissue/blood/brain expression, subcellular localization, cancer prognostic markers, batch lookup, antibody validation/staining); stdio, install from source (source).

Flagged

None this slot.

Verified (no changes)

No entries are older than 30 days; nothing required re-verification this slot.

2026-06-13

Chemistry directed pass plus a manifest sweep. The anthropics/life-sciences marketplace matched the catalogued set — every in-scope plugin/skill/connector is already a page. The Chemistry pass surfaced two genuinely-new installable cheminformatics MCP servers (ChemLint, ChemCP); MoziChem-MCP (chemical-engineering thermodynamics) and the still-missing retrosynthesis wrapper were deferred.

Added

ChemLint (Categories: Chemistry, Drug Repurposing and Discovery) — MIT; molML/ChemLint MCP server exposing 150+ molecular machine-learning tools (SMILES cleaning, descriptors/fingerprints, scaffolds, similarity/clustering, 33+ ML algorithms, stats, activity cliffs, dim-reduction); stdio, uv-launched (source).
ChemCP (Categories: Chemistry) — scottmreed/ChemCP MCP App rendering interactive 2D molecular structures from SMILES via RDKit.js with basic property readouts; npm install, stdio for Desktop / HTTP+tunnel for Claude.ai. License undeclared upstream — flagged (source).

Updated

RDKit MCP Server (TandemAI) — re-verified; source live, last_verified 2026-05-20 → 2026-06-13 (no field changes).

Flagged

ChemCP — no LICENSE file declared upstream; Free / OSS claim and redistribution terms marked Unverified — pending an explicit license.

Verified (no changes)

1 entry re-verified (rdkit-mcp).

2026-06-11

Two threads this date: (1) the daily directed pass on Molecular and Cellular Biology (Thursday focus) plus a manifest sweep and one user-request action; and (2) a batch ingest of two life-science skill collections following the K-Dense precedent (2026-06-04). For the batch, the K-Dense-only ingester was generalised into a reusable, collection-parameterised pipeline — scripts/ingest_collection.py driven by scripts/collections.yaml (per-collection registry) + an auditable scripts/<collection>_category_map.yaml (include = CREATE a new page, augment = add this collection’s install path to an existing page, skip = recorded out-of-scope). It handles flat and nested skills/ layouts and idempotent augmentation (a  sentinel). Catalog tool pages: 179 -> 338. The anthropics/life-sciences marketplace matched the catalogued set — every in-scope plugin/skill is already a page; biorxiv/clinical-trials stay flagged (mcp.deepsense.ai NXDOMAIN).

Added

SciAgent-Skills (jaechang-hits/SciAgent-Skills, 197 skills, CC BY 4.0, BixBench-evaluated) — 89 new pages: genomics (BWA-MEM2, STAR, Salmon, SAMtools/BCFtools, GATK, CNVkit, SnpEff, PLINK2, DESeq2, featureCounts, GSEApy, fastp, MultiQC, BEDtools, CellTypist, Harmony, popV) and ~24 reference-database skills (ENCODE, ClinVar, dbSNP, gnomAD, COSMIC, cBioPortal, GWAS Catalog, ClinPGx, JASPAR, ReMap, RegulomeDB, QuickGO, Monarch, KEGG, Reactome, STRING, BRENDA, ARCHS4, GEO, NCBI Gene, UCSC, Mouse Phenome Database, ZINC, openFDA, DailyMed, DDInter, Open Targets, GtoPdb, UniChem, EMDB, HMDB, Metabolomics Workbench, PRIDE, InterPro, MaxQuant); docking & MD (AutoDock Vina, smina, MDAnalysis, MDTraj); bio-imaging (Cellpose, napari, OpenCV, PyImageJ, scikit-image, trackpy); plus SAR analysis, ViennaRNA, pLannotate, sgRNA design, HOMER, MACS3, CellChat, libSBML, MOFA+, muon, SpikeInterface, nnU-Net, SimpleITK, Snakemake, Plotly, OpenAlex, bioRxiv, USPTO. 93 existing pages gained a SciAgent alternative install path (one entry per tool — incl. the Ensembl skill folded into the same-day ensembl MCP page). 15 out of scope — recorded in scripts/sciagent_category_map.yaml.
NeuroClaw (CUHK-AIM-Group/NeuroClaw, 86 skills, MIT) — 68 new Neuroscience pages: tool/modality wrappers (FreeSurfer, FSL, fMRIPrep, QSIPrep, CONN, DIPY, MNE-Python, Nilearn, NiBabel, dcm2niix, nii2dcm, NeuroHarmonize, WMH/ASL/PET/DWI/EEG/fMRI/sMRI/MEG, brain visualization), dataset pipelines (ADNI, UK Biobank, HCP-A/D/EP/YA, ABCD, ABIDE, ADHD-200, AIBL, AOMIC, BOLD5000, Cam-CAN, COBRE, HBN, IXI, NSD, OASIS, PNC, PPMI, REST-meta-MDD, SEED-IV/VIG, TCP, UCLA-CNP, NIFD, MND, MS-Challenge, DMT-HAR-MED), and phenotype-prediction model docs (BrainNetworkTransformer, BrainGNN, Com-BrainTF, IBGNN, LG-GNN, FM-APP, NeuroSTORM, GLM, ICA, K-means, SVM, SpaceNet, dictionary learning, detrending, filtering, hierarchical). 1 augment (bids). 17 skipped — recorded in scripts/neuroclaw_category_map.yaml.
Biomni (Categories: All) — Stanford SNAP Lab general-purpose biomedical AI agent; Claude Code skill via npx skills add … --skill biomni driving the biomni PyPI package (~11GB data lake). Apache-2.0; some bundled tools more restrictive (source).
Ensembl MCP Server (Categories: Drug Repurposing and Discovery, Immunology and Microbiology, Integrative Structural and Computational Biology, Molecular and Cellular Biology, Neuroscience, Translational Medicine) — MIT; npm ensembl-mcp-server, 10 read-only tools over the Ensembl REST API (source). The SciAgent Ensembl REST skill was folded into this page as an alternative install path rather than created as a duplicate.

Updated

93 existing tool pages augmented with a SciAgent (92) or NeuroClaw (1) alternative install path. AGENT.md community-collections table + batch-ingest note updated; scripts/ingest_collection.py + scripts/collections.yaml added; SciAgent and NeuroClaw moved to diff-only mode. curator-state.md: ZINC, Ensembl/UCSC MCP, and mne-neurophysiology deferred entries updated to reflect now-catalogued coverage.
10x Genomics Cloud MCP — user field report (#26, “worked great”) confirming the Claude Code plugin-marketplace install path; added a dated note in Notes, last_verified -> 2026-06-11.

Flagged

OpenClaw-Medical-Skills (FreedomIntelligence/OpenClaw-Medical-Skills) deferred on a license contradiction — README markets it as open-source, but there is no repo LICENSE file and 185 of 449 SKILL.md files carry an “All Rights Reserved … unauthorized copying strictly prohibited” proprietary header. Not ingested; see curator-state.md. Most of its life-science surface is already covered by the cleanly-licensed K-Dense/SciAgent ingests.

Verified (no changes)

anthropics/life-sciences marketplace diffed — no new plugins/skills. All 13 entries (PubMed, BioRender, Synapse, Scholar Gateway, Consensus, Cortellis, AdisInsight, 10x Genomics, single-cell-rna-qc, instrument-data-to-allotrope, nextflow-development, scvi-tools, scientific-problem-selection) already catalogued.

2026-06-09

Directed pass on Immunology and Microbiology (Tuesday focus) plus a manifest sweep. WebFetch was unavailable from the runner this entire run (every call returned a backend model 404 — claude-3-5-haiku not found), and all raw.githubusercontent.com/github.com/api.github.com marketplace.json fetches returned 404, so the marketplace diff and candidate verification leaned on corroborated WebSearch results only. The directed-pass seed queries (IEDB epitope MCP, antibody-design skill, BCR/TCR repertoire MCP, metagenomics/microbiome MCP) reconfirmed the three previously-logged gaps: no Claude-installable wrapper exists for IEDB’s PostgREST Query API, for AIRR-seq tooling (Immcantation / nf-core/airrflow / immunarch / immuneML), or for metagenomics pipelines (QIIME2 / Kraken2 / MetaPhlAn). The anthropics/life-sciences marketplace contents matched the catalogued set (no new plugins), and a K-Dense diff via search showed the collection steady at ~140/142 skills with no net-new immunology/microbiology directories. No new candidate could be verified to the followable-verbatim install standard from a primary source, so no tool pages were created. No last_verified dates exceed the 30-day window (oldest is 2026-05-19, 21 days).

Added

None.

Updated

None.

Flagged

None.

Verified (no changes)

Immunology/Microbiology-relevant entries reviewed against today’s seed queries; install paths and capability descriptions remain accurate. IEDB, AIRR-seq (BCR/TCR repertoire), and metagenomics/microbiome confirmed still uncovered by any Claude-installable wrapper — deferral dates refreshed in curator-state.md.

2026-06-08

Directed pass on Chemistry (Monday focus) plus a manifest sweep. WebFetch was unavailable from the runner this entire run (every call returned a backend model 404 — claude-3-5-haiku not found), and all raw.githubusercontent.com/api.github.com marketplace.json fetches returned 404, so the marketplace diff and candidate verification leaned on corroborated WebSearch results only. Because no new candidate could be verified to the followable-verbatim install standard from a primary source this run, no new tool pages were created. The Chemistry surface is already broad (rdkit-mcp, rdkit-skill, pubchem, chembl, datamol, molfeat, medchem, deepchem, diffdock, rowan, matchms, pytdc, torchdrug, molecule-mcp, pyopenms). The directed-pass seed queries (RDKit MCP, retrosynthesis MCP, ChEMBL MCP, Polaris, PubChem MCP) surfaced one genuinely new install path — a hosted/typescript PubChem server (cyanheads/pubchem-mcp-server) with an 8-tool read-only surface and a public Streamable HTTP endpoint — which was folded into the existing pubchem entry per the one-entry-per-tool rule. Two Chemistry candidates were deferred: a ZINC Database Claude Code skill (davila7/claude-code-templates) whose exact install flag and SKILL.md tool list could not be verified with WebFetch down, and retrosynthesis wrappers (AiZynthFinder/ASKCOS/IBM RXN), still without a Claude-installable surface.

Updated

PubChem MCP Server — added the cyanheads/pubchem-mcp-server install paths (hosted Streamable HTTP https://pubchem.caseyjhand.com/mcp via claude mcp add --transport http; mcp-remote proxy for Claude Desktop; local bunx stdio) and documented its wider 8-tool surface — substructure/superstructure/2D-similarity search and GHS hazard classification — alongside the existing JackKuo666 server; last_verified 2026-05-20 → 2026-06-08 (cyanheads/pubchem-mcp-server, npm).

Flagged

None.

Verified (no changes)

Chemistry-tagged entries (rdkit-mcp, chembl, datamol, molfeat, medchem, deepchem, diffdock, rowan, matchms, pytdc, torchdrug) reviewed against today’s seed queries — install paths and capability descriptions remain accurate. Retrosynthesis confirmed still uncovered by any Claude-installable wrapper.

2026-06-07

Directed pass on Drug Repurposing and Discovery (Sunday focus) plus a manifest sweep of anthropics/life-sciences. WebFetch and curl were both unavailable from the runner this run (WebFetch returned a backend model 404; raw marketplace.json fetch was not permitted), so the marketplace diff and candidate verification leaned on corroborated WebSearch results plus the Clarivate press release and the existing first-party plugin pattern (adisinsight, open-targets). Surfaced two first-party plugins from the deferred queue — Cortellis (Clarivate regulatory/pipeline intelligence) and Consensus (literature evidence synthesis) — both confirmed as life-sciences marketplace plugins with subscription/account gating. The directed-pass seed queries (Open Targets MCP, DrugBank MCP, drug-repurposing/target-prioritization MCP) surfaced only tools already catalogued (open-targets, chembl, drugbank, pubchem); no new standalone discovery MCP warranted an entry. No last_verified dates exceed the 30-day window (oldest is 2026-05-19, 19 days).

Added

Cortellis Plugin (Categories: Drug Repurposing and Discovery, Translational Medicine) — Clarivate Cortellis MCP plugin in anthropics/life-sciences; global drug-pipeline, clinical-trial, regulatory, safety, and deals intelligence; subscription-gated, free install (anthropics/life-sciences, Clarivate).
Consensus Plugin (Categories: All) — Consensus.app MCP plugin in anthropics/life-sciences; AI-powered scientific literature search and evidence synthesis; free Consensus account required (anthropics/life-sciences, Consensus).

Verified (no changes)

178 tool pages spot-checked for last_verified recency; all current within the 30-day window. Drug-discovery-tagged entries (open-targets, chembl, drugbank, pubchem, adisinsight, owkin) reviewed against today’s seed queries — install paths and capability descriptions remain accurate.

2026-06-06

Directed pass on Translational Medicine (Saturday focus) plus a manifest sweep. WebFetch was unreliable this run (the anthropics/life-sciences raw marketplace.json, GitHub READMEs, npm, and the MCP Registry API all returned backend 404/403 to automated fetches), so candidate verification leaned on corroborated WebSearch results across the npm listing, Glama, and the author’s own write-ups. Surfaced one fully-verifiable Translational entry from the deferred queue (@ythalorossy/openfda). The AACT Clinical Trials MCP (navisbio) was re-investigated but held back — its license is reported inconsistently (MIT vs GPLv3) and the canonical repo/package path is ambiguous (aact_mcp vs mcp-server-aact); deferred pending a primary-source license confirmation. No last_verified dates exceed the 30-day window (oldest is 2026-05-19, 18 days).

Added

OpenFDA MCP Server (ythalorossy) (Categories: Drug Repurposing and Discovery, Translational Medicine) — MIT-licensed npm MCP server (@ythalorossy/openfda) exposing 7 openFDA drug tools (adverse events, safety/labeling, manufacturer, NDC resolution); fully copy-pasteable claude mcp add-json and Claude Desktop install paths (github, npm).

Verified (no changes)

176 tool pages spot-checked for last_verified recency; all current within the 30-day window.

2026-06-05

Directed pass on Neuroscience (Friday focus). Manifest sweep of anthropics/life-sciences shows no diff vs. 2026-06-04 — same 10-plugin set; biorxiv@life-sciences / clinical-trials@life-sciences remain DOA per upstream issue #42 (mcp.deepsense.ai NXDOMAIN). Directed-pass seed queries (Allen Brain Atlas MCP, NWB MCP, fMRI / fMRIPrep Claude skill, MNE-Python skill, spike sorting MCP, DANDI MCP) surfaced only candidates already catalogued — the credible Neuroscience surface is well covered: allenbrain (Allen Brain Atlas RMA MCP), neurosift (DANDI/OpenNeuro + NWB introspection + PyNWB docs search MCP), openneuro (OpenNeuro GraphQL MCP), aind-data (AIND Metadata MCP + NWB file access), neurokit2 (Claude Skill — biosignals, with MNE integration in references/eeg.md), neuropixels-analysis (Claude Skill — Kilosort4 / SpykingCircus2 / Mountainsort5 spike-sorting pipeline incl. AI-assisted curation), and bids (Claude Skill — BIDS spec + fMRIPrep / MRIQC / QSIPrep BIDS-Apps). Two listings investigated and not added: an mne-neurophysiology-analysis MCP-market entry without a verifiable upstream GitHub source, and bendichter/dandi-query-mcp whose advertised tool surface is the boilerplate note-server template (create_note / summarize_notes), not a real DANDI wrapper — and DANDI semantic search is already covered by neurosift. No new entries this run; no last_verified dates older than 30 days (oldest is 2026-05-19, 17 days).

Verified (no changes)

176 tool pages spot-checked for last_verified recency; all current within the 30-day window. The Neuroscience-tagged subset (7 pages) reviewed against today’s seed queries — all install paths and capability descriptions remain accurate.

2026-06-04 — General-Purpose Utilities category

Added an 8th catalog category, General-Purpose Utilities, and ingested the domain-agnostic remainder of the K-Dense collection into it. Recipes assemble horizontal tooling (plotting, dataframes, ML/stats, scientific communication) alongside domain tools, so these belong in the catalog even though they aren’t life-science-specific. The new category is a cross-cutting utilities shelf, not a research area: tools tagged General-Purpose Utilities appear only on the utilities index, never on the seven research-area pages, and never carry the All tag. 68 previously-skipped K-Dense skills moved in (including consciousness-council, a multi-perspective deliberation utility); only 3 remain out of scope (dhdna-profiler, market-research-reports, paperzilla — novelty / business / product-locked, not reusable utilities). Catalog now totals 175 tool pages (after merging the concurrent daily curator run that had independently surfaced COBRApy / DiffDock / Adaptyv / DeepChem / PyTDC).

Added

New category index catalog/general-purpose-utilities.html (nav_order 8). AGENT.md, root README.md, and catalog/README.md updated to describe the seven research areas plus the utilities shelf; RECIPE_AGENT.md clarified that recipes use only the seven research-area subject areas (utilities are ingredients, not a problem domain).
67 General-Purpose Utilities skill pages (K-Dense), including: data/compute — polars, dask, vaex, zarr-python, networkx, modal, optimize-for-gpu; viz/communication — matplotlib, seaborn, scientific-visualization, scientific-schematics, scientific-writing, scientific-slides, latex-posters/pptx-posters, infographics, generate-image, markdown-mermaid-writing, markitdown, liteparse, pdf/docx/pptx/xlsx, citation-management, pyzotero, venue-templates, peer-review, scholar-evaluation, research-grants; ML/stats — scikit-learn, statsmodels, pymc, shap, umap-learn, sympy, pytorch-lightning, torch-geometric, transformers, stable-baselines3, pufferlib, pymoo, aeon, timesfm-forecasting; literature/web — paper-style exa-search, parallel-web, research-lookup, open-notebook, bgpt-style; adjacent-domain science — qiskit, cirq, pennylane, qutip, pymatgen, astropy, geopandas, geomaster, fluidsim, matlab; misc utilities — exploratory-data-analysis, statistical-analysis, hypogenic, hugging-science, autoskill, get-available-resources, what-if-oracle, usfiscaldata, consciousness-council.

Updated

scripts/kdense_category_map.yaml — 66 entries moved from skip to include tagged [General-Purpose Utilities], parallel-web added; skip trimmed to the 4 genuine non-utilities with reasons.
scripts/ingest_kdense.py — PRETTY display-name map extended for the 67 utility skills.

Verified (no changes)

All 173 tool pages pass front-matter lint; the seven research-area card counts are unchanged (utilities are tagged on the shelf only and do not appear on research-area pages); every General-Purpose Utilities tool is tagged on the shelf alone.

2026-06-04

One-time batch ingestion of the K-Dense scientific-agent-skills collection. Rather than continue surfacing K-Dense skills one or two at a time under the ≤5/run soft cap (the Deferred queue had grown a long K-Dense backlog), this run ingests the whole life-science-relevant subset in a single auditable pass and switches the K-Dense source to diff-only mode for future runs. The collection has 143 skills; 69 are life-science-relevant and are now catalogued (51 new pages + 18 existing K-Dense pages repaired). The remaining ~74 are general-purpose (quantum computing, materials science, generic ML/stats/viz, office-document and web-search tooling, infra) and are out of scope — enumerated with reasons in scripts/kdense_category_map.yaml so future runs don’t re-evaluate them. Generated pages are schema-accurate but intentionally lean (built from each skill’s SKILL.md front-matter via scripts/ingest_kdense.py); subsequent daily runs enrich them. Also fixes a real upstream breakage: K-Dense migrated scientific-skills/<name>/ → skills/<name>/ and replaced the (non-existent) claude-scientific-skills plugin marketplace with npx skills add — every existing K-Dense page carried broken install steps.

Added

51 new K-Dense Claude Skill pages (source: K-Dense-AI/scientific-agent-skills):

Chemistry / Drug discovery — deepchem, diffdock, matchms, pytdc, rowan, torchdrug
Structural & computational biology — esm, pyopenms, adaptyv
Molecular & cellular biology / genomics — biopython, bioservices, bulk-rnaseq, cobrapy, deeptools, etetoolkit, geniml, gtars, lamindb, pathway-enrichment, phylogenetics, polars-bio, pysam, tiledbvcf, primekg
Lab platforms & data integrations — benchling-integration, dnanexus-integration, ginkgo-cloud-lab, labarchive-integration, latchbio-integration, omero-integration, opentrons-integration, protocolsio-integration, pylabrobot
Translational medicine / clinical / imaging — clinical-decision-support, clinical-reports, treatment-plans, pyhealth, pydicom, histolab, pathml, imaging-data-commons, pacsomatic, scikit-survival, iso-13485-certification
Cross-cutting (All) — database-lookup, paper-lookup, literature-review, bgpt-paper-search, hypothesis-generation, scientific-brainstorming, scientific-critical-thinking

Updated

Install-path migration fix on 18 existing K-Dense pages — anndata, arboreto, bids, cellxgene-census, datamol, depmap, flowio, gget, glycoengineering, medchem, molecular-dynamics, molfeat, neurokit2, neuropixels-analysis, pydeseq2, rdkit-skill, scikit-bio, scvelo. Replaced the dead claude-scientific-skills plugin-marketplace block with npx skills add K-Dense-AI/scientific-agent-skills, migrated scientific-skills/ → skills/ clone paths, and bumped last_verified.
Three “one entry per tool” augments — added the K-Dense skill as an alternative install path on scanpy (existing scmcphub MCP), scvi-tools (existing Anthropic skill), and nextflow-development (existing Anthropic skill) rather than creating duplicate pages.
AGENT.md — K-Dense source switched to diff-only mode: fully ingested as of 2026-06-04; future runs diff skills/ against catalogued K-Dense slugs and add only new skills / flag removed ones, instead of incrementally surfacing the backlog.
catalog/curator-state.md — cleared the now-ingested K-Dense items from Deferred; Recently surfaced collapsed to a single batch line.
New pipeline files — scripts/kdense_category_map.yaml (auditable scope filter + category map) and scripts/ingest_kdense.py (the generator/repairer), reusable for future collection batches (SciAgent-Skills, OpenClaw-Medical-Skills, NeuroClaw).

Verified (no changes)

All 106 tool pages pass front-matter lint (required fields, ≤25-word summaries, canonical categories, slug-matched feedback footers).
2026-06-04

Directed pass on Molecular and Cellular Biology (Thursday focus). Manifest sweep of anthropics/life-sciences shows no diff vs. 2026-06-03 — same 10-plugin set (pubmed, biorender, synapse, wiley-scholar-gateway, 10x-genomics, single-cell-rna-qc, instrument-data-to-allotrope, nextflow-development, scvi-tools, scientific-problem-selection); biorxiv@life-sciences / clinical-trials@life-sciences remain DOA per upstream issue #42 (mcp.deepsense.ai NXDOMAIN). The directed pass surfaced COBRApy (Claude Skill) from the standing Deferred queue — the K-Dense skill wrapper around opencobra/cobrapy, the canonical Python library for constraint-based reconstruction and analysis of genome-scale metabolic models. COBRApy fills the metabolism slice that none of the existing Molecular and Cellular Biology entries cover: scanpy / scvelo / arboreto / pydeseq2 / scvi-tools / cellxgene-census are transcriptomics, gget / uniprot are sequence/identifier resolvers, and single-cell-rna-qc is QC — none drive FBA / FVA / knockout screens / flux sampling on SBML models. One new entry, well under the 5-entry soft cap; no user-feedback issues in the open queue this run.

Added

COBRApy (Claude Skill) (Categories: Molecular and Cellular Biology, Chemistry, Drug Repurposing and Discovery) — K-Dense skill driving COBRApy for constraint-based metabolic modelling. Loads SBML / JSON / YAML genome-scale models; runs flux balance analysis (FBA, pFBA, geometric FBA), flux variability analysis (with loopless option and fraction_of_optimum control), single and double gene/reaction knockouts, Markov-chain flux sampling, production envelopes, and gapfilling. Solver backends pluggable via model.solver (GLPK default; CPLEX / Gurobi for large models). Skill is MIT-licensed (collection); COBRApy itself is GPL-2.0 — review for commercial use (source, opencobra/cobrapy, Ebrahim et al. BMC Syst. Biol. 2013, Playbooks: cobrapy skill).

Updated

catalog/curator-state.md — Recently surfaced refreshed with the COBRApy addition (oldest of the prior five — npi-registry — rolled off; diffdock, adaptyv, deepchem, pytdc carried forward). Deferred queue: removed cobrapy from the TorchDrug / PyOpenMS / matchms / cobrapy chemistry-stack-siblings line (cobrapy now surfaced); annotated the residual line with promotion dates and what each remaining sibling covers (TorchDrug = protein representation learning; PyOpenMS = MS proteomics; matchms = MS/MS spectral matching).

Verified (no changes)

Existing Molecular and Cellular Biology entries (scanpy last_verified 2026-05-26, anndata 2026-05-26, scvi-tools 2026-05-22, pydeseq2 2026-05-21, scvelo 2026-05-28, arboreto 2026-05-28, cellxgene-census 2026-05-29, single-cell-rna-qc 2026-05-22, plus cross-cutting All-tagged tools gget, uniprot, pubmed, biomcp, biocontextai) all within the 30-day verification window — no scheduled re-verification needed this run.
Manifest sweep of anthropics/life-sciences re-confirmed: no diff vs. 2026-06-03; biorxiv / clinical-trials plugins still DOA per upstream issue #42 (mcp.deepsense.ai NXDOMAIN). Deferred queue (Cortellis, Medidata, Consensus, Augmented-Nature ChEMBL, AACT, OpenFDA / Azure FHIR, NeuroClaw, MCP_Vina, TorchDrug / PyOpenMS / matchms, IEDB / Immcantation / metagenomics, Cryo-EM, RFdiffusion / ProteinMPNN, Rowan, Boltz / Protein Hunter) carries forward.

User requests

## User requests (open) was empty entering this run — nothing to action; section remains _None._.

2026-06-03

Directed pass on Integrative Structural and Computational Biology (Wednesday focus). Manifest sweep of anthropics/life-sciences shows no diff vs. 2026-06-02 — same 10-plugin set (pubmed, biorender, synapse, wiley-scholar-gateway, 10x-genomics, single-cell-rna-qc, instrument-data-to-allotrope, nextflow-development, scvi-tools, scientific-problem-selection); biorxiv@life-sciences / clinical-trials@life-sciences remain DOA per upstream issue #42 (mcp.deepsense.ai NXDOMAIN). The directed pass surfaced DiffDock (Claude Skill) from the standing Deferred queue — the K-Dense skill wrapper around gcorso/DiffDock, a diffusion-generative docker that predicts protein–ligand binding poses from a PDB structure plus a ligand SMILES, without exhaustive grid search. DiffDock complements the existing Structural-bio set: alphafold and pdb supply the receptor structure, molecular-dynamics refines the predicted poses, molecule-mcp visualises them in PyMOL/ChimeraX, and deepchem / medchem / pytdc re-score with affinity models. The Structural-bio pass also re-checked Cryo-EM (RELION / cryoSPARC / CTFFIND), RFdiffusion / ProteinMPNN standalone wrappers, Rowan MCP (k-yenko/rowan-mcp), and the Boltz / Protein Hunter MCPs (longevity-genie/protein_hunter_mcp, biomolecular-design-nexus/boltz_mcp) — none have a copy-pasteable Claude-installable path today, all now in the Deferred queue with the date stamped 2026-06-03. One new entry, well under the 5-entry soft cap; no user-feedback issues in the open queue this run.

Added

DiffDock (Claude Skill) (Categories: Drug Repurposing and Discovery, Integrative Structural and Computational Biology) — K-Dense skill driving DiffDock for diffusion-based protein–ligand docking. Generates N diffusion-sampled binding poses (default 10–40) with per-pose confidence scores; bundled scripts (prepare_batch_csv.py, analyze_results.py, setup_check.py) and references (confidence_and_limitations.md, parameters_reference.md, workflows_examples.md) for virtual screening, lead optimisation, and blind docking against AlphaFold-predicted targets. Poses only — pair with GNINA / Vina / MM-GBSA / FEP for affinity. MIT-licensed skill, MIT-licensed model (source, DiffDock — gcorso/DiffDock, Corso et al. ICLR 2023, Playbooks: diffdock skill).

Updated

catalog/curator-state.md — Recently surfaced refreshed with the DiffDock addition (oldest of the prior five — openneuro — rolled off; adaptyv, deepchem, pytdc, npi-registry carried forward). Deferred queue: removed DiffDock from the TorchDrug / DiffDock / PyOpenMS / matchms / cobrapy chemistry-stack-siblings line (DiffDock now surfaced); bumped the Cryo-EM and RFdiffusion / ProteinMPNN re-check dates to 2026-06-03 and noted that RFdiffusion / ProteinMPNN ship bundled inside longevity-genie/protein_hunter_mcp; added two new Structural-pass deferred lines — Rowan MCP (k-yenko/rowan-mcp) and Boltz / Protein Hunter MCP.

Verified (no changes)

Existing Integrative Structural and Computational Biology entries (alphafold last_verified 2026-05-19, pdb 2026-05-19, molecular-dynamics 2026-05-27, molecule-mcp 2026-05-20, plus the cross-cutting All-tagged tools gget, uniprot, pubmed, biomcp, biocontextai) all within the 30-day verification window — no scheduled re-verification needed this run.
Manifest sweep of anthropics/life-sciences re-confirmed: no diff vs. 2026-06-02; biorxiv / clinical-trials plugins still DOA per upstream issue #42 (mcp.deepsense.ai NXDOMAIN). Deferred queue (Cortellis, Medidata, Consensus, Augmented-Nature ChEMBL, AACT, OpenFDA / Azure FHIR, NeuroClaw, MCP_Vina, TorchDrug / PyOpenMS / matchms / cobrapy, IEDB / Immcantation / metagenomics, Cryo-EM, RFdiffusion / ProteinMPNN, Rowan, Boltz / Protein Hunter) carries forward.

User requests

## User requests (open) was empty entering this run — nothing to action; section remains _None._.

2026-06-02

Directed pass on Immunology and Microbiology (Tuesday focus). Manifest sweep of anthropics/life-sciences shows no diff vs. 2026-06-01 — same 10-plugin set (pubmed, biorender, synapse, wiley-scholar-gateway, 10x-genomics, single-cell-rna-qc, instrument-data-to-allotrope, nextflow-development, scvi-tools, scientific-problem-selection), and biorxiv@life-sciences / clinical-trials@life-sciences remain DOA per upstream issue #42 (mcp.deepsense.ai NXDOMAIN). The directed pass surfaced Adaptyv (Claude Skill) from the standing Deferred queue — the K-Dense skill that closes the design–build–test loop on the Adaptyv cloud lab, pre-screening sequences in silico (NetSolP / SoluProt / SolubleMPNN / ESM / ipTM / pSAE) and submitting them to wet-lab binding (BLI K_D / k_on / k_off), expression (E. coli / mammalian / yeast / insect), thermostability (DSF / CD), and enzyme-activity assays. Adaptyv is the first Immunology-tagged entry covering the actual antibody-design → wet-lab handoff (the existing glycoengineering entry covers glycan-shield analysis but not the design-build-test loop). The Immunology pass also re-checked the IEDB Query API, Immcantation / nf-core/airrflow, and metagenomics (QIIME2 / Kraken2 / MetaPhlAn): none have a Claude-installable wrapper as of today’s WebSearch — all three remain in the Deferred queue with the date bumped to 2026-06-02. One new entry, well under the 5-entry soft cap; no user-feedback issues in the open queue this run.

Added

Adaptyv (Claude Skill) (Categories: Immunology and Microbiology, Drug Repurposing and Discovery, Integrative Structural and Computational Biology) — K-Dense skill wrapping the Adaptyv cloud-lab API for closed-loop protein and antibody design. In-silico pre-screening with NetSolP / SoluProt (solubility), SolubleMPNN (expression-aware redesign), ESM (sequence-likelihood scoring), ipTM (AlphaFold-Multimer interface stability), and pSAE (aggregation risk); wet-lab assay submission for binding (biolayer interferometry K_D / k_on / k_off), expression in four cell systems, thermostability (DSF / CD T_m), and enzyme activity. Supports batch submission and webhook callbacks; turnaround ~21 days; the skill writes to a wet lab so review submissions carefully. Published via the K-Dense-AI/claude-scientific-skills plugin marketplace (source, scientific-skills/adaptyv/reference/examples.md, Adaptyv Bio, Playbooks: adaptyv skill).

Updated

catalog/curator-state.md — Recently surfaced refreshed with the Adaptyv addition (oldest of the prior five — cellxgene-census — rolled off; deepchem, pytdc, npi-registry, openneuro carried forward). Deferred queue: removed Adaptyv (K-Dense Skill) (surfaced this run); bumped the IEDB MCP wrapper and BCR/TCR repertoire MCP re-check dates to 2026-06-02 and expanded the latter to call out Immcantation 4.7.0-2026.01.21 / nf-core/airrflow / VDJdb / McPAS-TCR specifically; added a new Metagenomics / microbiome MCP deferred line (no Claude-installable wrapper for QIIME2 / Kraken2 / MetaPhlAn on the 2026-06-02 sweep).

Verified (no changes)

Existing Immunology and Microbiology entries (glycoengineering last_verified 2026-05-26, plus all All-tagged cross-cutting tools — pubmed, uniprot, alphafold, pdb, gget, biomcp, anndata, scanpy, scvi-tools, scikit-bio, flowio, 10x-genomics-cloud, pydeseq2, single-cell-rna-qc, nextflow-development, instrument-data-to-allotrope, cellxgene-census) all within the 30-day verification window — no scheduled re-verification needed this run.
Manifest sweep of anthropics/life-sciences re-confirmed: no diff vs. 2026-06-01; biorxiv / clinical-trials plugins still DOA per upstream issue #42 (mcp.deepsense.ai NXDOMAIN). Deferred queue (Cortellis, Medidata, Consensus, Augmented-Nature ChEMBL, AACT, OpenFDA / Azure FHIR, NeuroClaw, MCP_Vina, DiffDock / cobrapy / TorchDrug / PyOpenMS / matchms, IEDB / Immcantation / metagenomics) carries forward.

User requests

## User requests (open) was empty entering this run — nothing to action; section remains _None._.

2026-06-01

Directed pass on Chemistry (Monday focus). Manifest sweep of anthropics/life-sciences shows no diff vs. 2026-05-31 — the DOA biorxiv@life-sciences / clinical-trials@life-sciences plugins (upstream issue #42, mcp.deepsense.ai NXDOMAIN) still un-restored; no other new plugins this week. The directed pass surfaced DeepChem from the standing Deferred queue — the K-Dense skill wrapper around the canonical molecular-machine-learning framework (deepchem/deepchem), providing GCN / GAT / MPNN / AttentiveFP graph neural nets, molecular featurization, and the MoleculeNet benchmark suite (Tox21, ClinTox, SIDER, BBBP, QM7/8/9, ESOL, FreeSolv, Lipophilicity). DeepChem complements the existing rdkit-skill, medchem, datamol, molfeat, and pytdc Chemistry entries by adding the model-training and MoleculeNet-benchmark layer that none of those cover (RDKit / MedChem are featurizers and filters; PyTDC is benchmark splits and oracles). The Chemistry pass also re-checked retrosynthesis and docking MCPs: MCP_Vina (shogo-d-nakamura/MCP_Vina) is a viable AutoDock Vina MCP server but currently supports only AKT1 as a target — added to Deferred rather than catalogued. AiZynthFinder / ASKCOS / IBM RXN still lack any Claude-installable wrapper as of today’s WebSearch. One new entry, well under the 5-entry soft cap; no user-feedback issues in the open queue this run.

Added

DeepChem (Claude Skill) (Categories: Chemistry, Drug Repurposing and Discovery) — K-Dense skill driving DeepChem for molecular machine learning. Provides graph neural networks (GCN, GAT, MPNN, AttentiveFP), molecular featurization (graphs, Morgan / circular fingerprints, 2D/3D descriptors), pre-trained molecular foundation models, and the MoleculeNet benchmark suite (50+ datasets across quantum-chemistry, physical-chemistry, biophysics, and physiology / toxicity). TensorFlow and PyTorch backends; integrates with RDKit (source, DeepChem, deepchem/deepchem, MoleculeNet, Wu et al. Chem. Sci. 2018).

Updated

catalog/curator-state.md — Recently surfaced refreshed with the DeepChem addition (oldest of the prior five — scvelo — rolled off; pytdc, npi-registry, openneuro, cellxgene-census carried forward). Deferred queue updated: removed DeepChem (K-Dense Skill) (surfaced this run); annotated TorchDrug / DiffDock / PyOpenMS / matchms / cobrapy line to note the post-DeepChem residual; added MCP_Vina (AutoDock Vina MCP server, single-target AKT1) as a deferred Chemistry candidate.

Verified (no changes)

Existing Chemistry entries (rdkit-skill last_verified 2026-05-20, rdkit-mcp 2026-05-20, chembl 2026-05-30, pubchem 2026-05-20, medchem 2026-05-25, datamol 2026-05-25, molfeat 2026-05-25, molecule-mcp 2026-05-20, drugbank 2026-05-20, pytdc 2026-05-31, instrument-data-to-allotrope 2026-05-19) all within the 30-day verification window — no scheduled re-verification needed this run.
Manifest sweep of anthropics/life-sciences re-confirmed: no diff vs. 2026-05-31; biorxiv / clinical-trials plugins still DOA per upstream issue #42 (mcp.deepsense.ai NXDOMAIN). Deferred queue (Cortellis, Medidata, Consensus, Augmented-Nature ChEMBL, AACT, OpenFDA / Azure FHIR / TorchDrug / DiffDock / PyOpenMS / matchms / cobrapy / MCP_Vina) carries forward.

User requests

## User requests (open) was empty entering this run — nothing to action; section remains _None._.

2026-05-31

Directed pass on Drug Repurposing and Discovery (Sunday focus). Manifest sweep of anthropics/life-sciences shows no diff vs. 2026-05-30 — the DOA biorxiv@life-sciences / clinical-trials@life-sciences plugins (upstream issue #42, mcp.deepsense.ai NXDOMAIN) still un-restored; no other new plugins this week. The directed pass surfaced PyTDC from the standing Deferred queue — the K-Dense skill wrapper around Therapeutics Data Commons (Harvard mims-harvard/TDC), the canonical benchmark suite for ADMET, drug-target / drug-drug interactions, drug response, molecular generation, and retrosynthesis. PyTDC complements the existing chembl, drugbank, open-targets, pubchem, medchem, datamol, molfeat, and rdkit-skill Drug Repurposing entries by providing the labelled benchmark splits and generation oracles those upstream tools then featurize / model. One new entry, well under the 5-entry soft cap; no user-feedback issues in the open queue this run.

Added

PyTDC (Claude Skill) (Categories: Drug Repurposing and Discovery, Chemistry, Translational Medicine) — K-Dense skill driving the PyTDC Python client for Therapeutics Data Commons benchmarks: single_pred (ADMET, toxicity), multi_pred (DTI, DDI, drug response, PPI), and generation (de-novo molecules with 17+ oracles incl. QED / SA / DRD2 / GSK3B / JNK3, plus retrosynthesis and reaction yield). Bundled scripts (load_and_split_data.py, benchmark_evaluation.py, molecular_generation.py) plus datasets.md / oracles.md / utilities.md references (source, TDC, mims-harvard/TDC, Huang et al. NeurIPS 2021).

Updated

catalog/curator-state.md — Recently surfaced refreshed with the PyTDC addition (oldest of the prior five — arboreto — rolled off; cellxgene-census, scvelo, openneuro, npi-registry carried forward). Deferred queue updated: removed PyTDC from the TorchDrug / PyTDC / DiffDock / PyOpenMS / matchms / cobrapy (K-Dense Skills) chemistry-stack-siblings line (PyTDC now surfaced).

Verified (no changes)

Existing Drug Repurposing and Discovery entries (open-targets last_verified 2026-05-20, drugbank 2026-05-20, chembl 2026-05-30, pubchem, medchem, datamol, molfeat, rdkit-skill, rdkit-mcp, molecule-mcp, adisinsight, owkin, depmap) all within the 30-day verification window — no scheduled re-verification needed this run.
Manifest sweep of anthropics/life-sciences re-confirmed: no diff vs. 2026-05-30; biorxiv / clinical-trials plugins still DOA per upstream issue #42 (mcp.deepsense.ai NXDOMAIN). Deferred queue (Cortellis, Medidata, Consensus, Augmented-Nature ChEMBL, AACT, OpenFDA / Azure FHIR / DeepChem / TorchDrug / DiffDock MCPs and skills) carries forward.

User requests

## User requests (open) was empty entering this run — nothing to action; section remains _None._.

2026-05-30

Directed pass on Translational Medicine (Saturday focus). Manifest sweep of anthropics/life-sciences shows no diff vs. 2026-05-29 — the DOA biorxiv@life-sciences / clinical-trials@life-sciences plugins still un-restored, no other new plugins this week. The directed pass cleared NPI Registry MCP from the standing Deferred queue (third Anthropic Healthcare MCP-marketplace entry alongside cms-coverage and icd-10-codes, plus the existing pubmed connector — completes the four Healthcare-marketplace MCPs published with the Jan 2026 Claude for Healthcare launch). Two user-feedback issues processed: #17 with a structured worked fine trailer for chembl, and #15 a no-trailer migration smoke-test against the same tool. One new entry, well under the 5-entry soft cap; both user requests closed this run.

Added

NPI Registry MCP (Anthropic Healthcare) (Categories: Translational Medicine) — Anthropic-published MCP server over the CMS NPPES NPI Registry API v2.1 with npi_validate / npi_lookup / npi_search tools for US healthcare-provider credential verification, network construction, and clinical-trial-investigator validation. Pairs with the existing cms-coverage, icd-10-codes, and prior-auth-review entries for end-to-end prior-auth review workflows (Anthropic tutorial, anthropics/healthcare, CMS NPPES).

Updated

chembl — added a dated field-report note in Notes (/plugin marketplace add anthropics/life-sciences + /plugin install chembl@life-sciences reported working without modification per user feedback issue #17); last_verified bumped 2026-05-24 → 2026-05-30.
catalog/curator-state.md — Recently surfaced refreshed with the NPI Registry addition (oldest of the prior five — molecular-dynamics — rolled off). Removed NPI Registry MCP (Anthropic Healthcare) from the Deferred queue (surfaced this run); added easysolutions906 Healthcare MCP (@easysolutions906/mcp-healthcare — 10-tool ICD-10 / NPI / NDC / DEA community bundle) to Deferred as a single-install alternative to the four discrete Anthropic Healthcare MCPs.

Verified (no changes)

Existing Translational Medicine entries (fhir-developer, prior-auth-review, fhir-wso2, cms-coverage, icd-10-codes, clinical-trial-protocol, clinicaltrials-gov-mcp) all within the 30-day verification window (oldest 2026-05-23) — no scheduled re-verification needed this run.
Manifest sweep of anthropics/life-sciences re-confirmed: no diff vs. 2026-05-29; biorxiv / clinical-trials plugins still DOA per upstream issue #42 (mcp.deepsense.ai NXDOMAIN). Deferred queue (Cortellis, Medidata, Consensus, Augmented-Nature ChEMBL, AACT, OpenFDA / Azure FHIR MCPs) carries forward.

User requests

#15 — [Tool feedback] migration smoke test — chembl with no tool-feedback trailer; closed this run with a note that the title-derived intent is a post-migration form-plumbing smoke test, no content action needed beyond the chembl refresh already driven by #17.
#17 — structured trailer feedback-on=chembl | sentiment=worked fine; closed this run by adding a field-report Notes line on the chembl page and refreshing last_verified.

2026-05-29

Directed pass on Neuroscience (Friday focus). Manifest sweep of anthropics/life-sciences shows no diff vs. 2026-05-28 — no new neuro plugins shipped upstream this week; the biorxiv@life-sciences / clinical-trials@life-sciences DOA situation continues. Existing Neuroscience entries (allenbrain, aind-data, neurosift, neurokit2, neuropixels-analysis, bids) all within the 30-day verification window — no re-verification needed. Directed search turned up two strong candidates: OpenNeuro MCP (QuentinCody) — a hosted Cloudflare Workers SSE server wrapping the OpenNeuro GraphQL API, with a clean copy-pasteable install path documented upstream — and NeuroClaw (CUHK-AIM Group), an 81-skill neuroimaging library with FreeSurfer / FSL / fMRIPrep / MNE / nilearn / DIPY integrations. OpenNeuro MCP was added; NeuroClaw was deferred because the upstream README positions skills/ as Claude-Code-installable but does not publish a copy-pasteable ~/.claude/skills/ snippet and the license terms could not be confirmed under today’s WebFetch reliability. One new entry, well under the 5-entry soft cap.

Added

OpenNeuro MCP (Categories: Neuroscience) — Community MCP server wrapping the OpenNeuro GraphQL API for MRI / MEG / EEG / iEEG / ECoG dataset, snapshot, and file-listing queries; hosted Cloudflare Workers SSE endpoint, MIT + Academic Citation Requirement license. Complements the Neurosift Tools MCP (DANDI + NWB) by covering OpenNeuro’s archive (source, OpenNeuro).

Updated

catalog/curator-state.md — Recently surfaced refreshed with this run’s OpenNeuro MCP addition (oldest of the prior five rolled off). Deferred queue gained NeuroClaw (install path gap + license unconfirmed) and a watch item on the K-Dense v2.43.0 scientific-skills/ → skills/ path migration that may affect future K-Dense skill page edits.
Feedback-footer URL backfill — five tool pages added on / since 2026-05-27 (openneuro, cellxgene-census, molecular-dynamics, scvelo, arboreto) still pointed their Installed this tool? feedback links at the pre-migration goodb/sci-ai-enabler org / goodb.github.io pages host. The org-rename commit (4d703fe) ran on a branch that forked before today’s earlier catalog commit (c102ee7), so these five files were not rewritten by the migration. Rewrote both occurrences in each footer to scripps-ai-enablement/sci-ai-enabler and scripps-ai-enablement.github.io to restore the working feedback path. No content changes outside the footer URLs (migration commit).

Verified (no changes)

All Neuroscience catalog entries’ last_verified is within the 30-day window (oldest 2026-05-20) — no scheduled re-verification needed this run.
Manifest sweep of anthropics/life-sciences re-confirmed: no new neuro plugins; biorxiv / clinical-trials plugins still DOA per upstream issue #42 (mcp.deepsense.ai NXDOMAIN). Deferred queue (Cortellis, Medidata, Consensus, NPI Registry, Augmented-Nature ChEMBL, AACT, OpenFDA / Azure FHIR MCPs) carries forward.

User requests

#11 — [Tool feedback] with no tool-feedback trailer; still unactionable from the curator runner (no gh CLI / GitHub-API auth in this environment). Closed this run with that note; inbound responder will re-surface if the issue is updated.

2026-05-28

Directed pass on Molecular and Cellular Biology (Thursday focus). Manifest sweep of anthropics/life-sciences shows no diff vs. 2026-05-27 sweep among plugins our catalog already covers; an e96556b upstream commit and issue #42 indicate that biorxiv@life-sciences and clinical-trials@life-sciences plugins are now published in the marketplace but currently DOA (the mcp.deepsense.ai host returns NXDOMAIN) — held off from cataloguing pending Anthropic / DeepSense restoring the endpoint, and added to the Deferred queue. The directed pass surfaced three K-Dense single-cell skills from the standing Deferred queue — Cellxgene Census, scVelo, and Arboreto — which collectively round out the catalog’s single-cell trajectory + GRN coverage that pairs with the existing scanpy, anndata, scvi-tools, and pydeseq2 entries. Three new entries, well under the 5-entry soft cap.

Added

Cellxgene Census (Claude Skill) (Categories: Molecular and Cellular Biology, Immunology and Microbiology, Translational Medicine) — K-Dense skill for querying the CZ CELLxGENE Discover census (50M+ cells, 1,000+ datasets) via TileDB-SOMA, with AnnData / Scanpy integration for reference-atlas construction and meta-analyses (source, CELLxGENE Census docs).
scVelo (Claude Skill) (Categories: Molecular and Cellular Biology, Neuroscience) — K-Dense skill driving scVelo for RNA-velocity analysis: steady-state / stochastic / dynamical models, latent-time inference, driver-gene identification, and velocity-embedding projection on UMAP / t-SNE (source, scVelo docs).
Arboreto (Claude Skill) (Categories: Molecular and Cellular Biology, Drug Repurposing and Discovery) — K-Dense skill for gene-regulatory-network inference with GRNBoost2 / GENIE3, scaled with Dask; standard upstream step for pySCENIC pipelines and regulator prioritisation (source, Arboreto docs).

Updated

catalog/curator-state.md — Recently surfaced refreshed with this run’s three additions; Deferred queue trimmed (removed Cellxgene Census, scVelo, Arboreto) and extended with the upstream-broken biorxiv@life-sciences and clinical-trials@life-sciences plugins.

Flagged

biorxiv@life-sciences and clinical-trials@life-sciences plugins — published in anthropics/life-sciences marketplace but the backing mcp.deepsense.ai MCP host returns NXDOMAIN as of upstream commit e96556b (2026-05-12). Not yet added to the catalog because there is no working install path to document. Watch upstream and revisit on next sweep.

Verified (no changes)

All catalog entries’ last_verified is within the 30-day window (oldest 2026-05-19); no scheduled re-verification needed this run.
Manifest sweep of anthropics/life-sciences re-confirmed: catalogued plugins (PubMed, BioRender, Synapse, Wiley Scholar Gateway, 10x Genomics, AdisInsight, single-cell-rna-qc, instrument-data-to-allotrope, nextflow-development, scvi-tools, scientific-problem-selection) still present; other deferred-queue plugins (Cortellis, Medidata, Consensus, NPI Registry) carry forward.

User requests

#11 — [Tool feedback] with no tool-feedback trailer; the curator runner cannot fetch issue bodies via the GitHub REST API or gh CLI in this environment (auth-gated), so the request stays in ## User requests (open) for the next run that has issue-body fetch capability. Closed-this-run note records the access constraint.

2026-05-27

Directed pass on Integrative Structural and Computational Biology (Wednesday focus). Manifest sweep of anthropics/life-sciences re-confirmed no diff vs. 2026-05-26 — every plugin currently published there remains catalogued. The K-Dense claude-scientific-skills catalog was scanned for structural-biology skills not yet in the catalog; Molecular Dynamics (OpenMM + MDAnalysis) is the strongest fit and complements the existing alphafold / pdb / molecule-mcp entries by covering the simulation and trajectory-analysis layer that none of those cover. Focused searches for cryo-EM (RELION / cryoSPARC) and protein-design model (RFdiffusion / ProteinMPNN) Claude wrappers turned up nothing installable today — deferred for the next Structural-focus pass. One new entry this run, well under the 5-entry soft cap.

Added

Molecular Dynamics (Claude Skill) (Categories: Integrative Structural and Computational Biology, Drug Repurposing and Discovery) — K-Dense skill that sets up, runs, and analyzes molecular-dynamics simulations end-to-end via OpenMM and MDAnalysis (system prep, force-field assignment, NVT/NPT equilibration, production MD, RMSD / RMSF / contact-map / free-energy-surface analysis) (source, K-Dense scientific-skills catalog).

Updated

catalog/curator-state.md — Recently surfaced refreshed with this run’s Molecular Dynamics addition; Deferred queue gained MDAnalysis (K-Dense Skill) (pending confirmation of whether it ships as a distinct SKILL.md), Cryo-EM MCP / Skill (no Claude-installable wrapper located this pass), and RFdiffusion / ProteinMPNN Claude Skill (no wrapper located this pass).

Verified (no changes)

Existing Integrative Structural and Computational Biology entries (alphafold, pdb, molecule-mcp, uniprot) spot-checked against upstream — install paths, supplier links, and pricing claims still valid; no field drift.
Manifest sweep of anthropics/life-sciences re-confirmed: no diff vs. 2026-05-26 sweep; deferred queue (Cortellis, Medidata, Consensus, NPI Registry, Augmented-Nature ChEMBL community alternative, AACT, OpenFDA / Azure FHIR MCPs) carries forward.

User requests

#11 — [Tool feedback] with no tool-feedback trailer emitted by the responder; issue body not fetchable from the curator runner this run, left open in ## User requests (open) for next-run triage.

2026-05-26

Directed pass on Immunology and Microbiology (Tuesday focus). Manifest sweep of anthropics/life-sciences shows no diff vs. 2026-05-25 — every plugin currently published there is already catalogued. The K-Dense claude-scientific-skills directory was scanned for immunology-relevant skills not yet in the catalog; Glycoengineering is the strongest fit (therapeutic-antibody and vaccine design via N/O-glycosylation prediction and external glyco-tool orchestration) and is the single new entry this run. A focused look for IEDB and BCR/TCR repertoire MCP servers turned up no Claude-installable wrappers — deferred for the next Immunology pass. Well under the 5-entry soft cap, by design: the directed pass is meant to surface high-fit category entries, not pad the catalog.

Added

Glycoengineering (Claude Skill) (Categories: Immunology and Microbiology, Drug Repurposing and Discovery) — K-Dense skill that scans protein sequences for N-glycosylation sequons (N-X-S/T), predicts O-glycosylation hotspots, and orchestrates NetOGlyc / GlycoShield / GlycoWorkbench for therapeutic-antibody Fc engineering, vaccine-immunogen design, and biologics developability (source, K-Dense scientific-skills catalog).

Updated

catalog/curator-state.md — Recently surfaced refreshed with this run’s Glycoengineering addition; Deferred queue gained IEDB MCP wrapper, BCR/TCR repertoire MCP, and Adaptyv (K-Dense Skill) as Immunology-focus candidates carried forward.

Verified (no changes)

Existing Immunology and Microbiology entries (scikit-bio, flowio, uniprot, gget, pydeseq2, nextflow-development) spot-checked against upstream — install paths, supplier links, and pricing claims still valid; no field drift.
Manifest sweep of anthropics/life-sciences re-confirmed: no diff vs. 2026-05-25 sweep; deferred queue (Cortellis, Medidata, Consensus, NPI Registry, Augmented-Nature ChEMBL community alternative) carries forward.

2026-05-25

Directed pass on Chemistry (Monday focus). Manifest sweep of anthropics/life-sciences shows no diff vs. 2026-05-24 — the marketplace’s chemistry surface (chembl, tooluniverse, plus the broader bio-research bundle) is fully catalogued. The directed pass picked up three sibling cheminformatics skills from the K-Dense claude-scientific-skills marketplace that round out the existing rdkit-skill entry into the full RDKit-derived lead-optimisation stack documented in K-Dense’s example workflows. Three new entries — at the soft cap minus two — with DeepChem and the remaining chemistry siblings (TorchDrug, PyTDC, DiffDock, PyOpenMS, matchms, cobrapy) carried forward on the deferred queue.

Added

Datamol (Claude Skill) (Categories: Chemistry, Drug Repurposing and Discovery) — K-Dense skill wrapping the Datamol RDKit-built library for molecular standardization, tautomer / stereoisomer enumeration, featurization, and parallel processing on large compound libraries (source, Datamol).
Molfeat (Claude Skill) (Categories: Chemistry, Drug Repurposing and Discovery) — K-Dense skill wrapping Molfeat, a unified API over 100+ molecular featurizers spanning classical fingerprints (ECFP, MACCS), 2D/3D descriptors, molecular graphs, and pretrained chemical foundation-model embeddings (ChemBERTa, MolBERT, Uni-Mol) (source, Molfeat docs).
MedChem (Claude Skill) (Categories: Chemistry, Drug Repurposing and Discovery) — K-Dense skill wrapping MedChem — drug-likeness filters (Lipinski, Veber, Egan, Muegge), PAINS / BRENK structural alerts, complexity metrics, and synthetic-accessibility scoring for hit-list triage (source, MedChem docs).

Updated

catalog/curator-state.md — Recently surfaced refreshed with this run’s three Chemistry additions; Deferred queue gained DeepChem, the remaining K-Dense chemistry-stack siblings (TorchDrug, PyTDC, DiffDock, PyOpenMS, matchms, cobrapy) as a bundle line, and a Retrosynthesis MCP entry (no Claude-installable wrapper for AiZynthFinder / ASKCOS / IBM RXN located on this pass).

Verified (no changes)

Existing Chemistry pages (rdkit-skill, rdkit-mcp, molecule-mcp, chembl, pubchem) spot-checked against upstream — install paths, supplier links, and pricing claims still valid; no field drift.
Manifest sweep of anthropics/life-sciences re-confirmed: no diff vs. 2026-05-24 sweep; deferred queue (Cortellis, Medidata, Consensus, NPI Registry, Augmented-Nature ChEMBL community alternative) carries forward.

2026-05-24

Directed pass on Drug Repurposing and Discovery (Sunday focus). The anthropics/life-sciences marketplace gained three first-party drug-discovery surfaces at J.P. Morgan 2026 — ChEMBL, ToolUniverse, and Owkin Pathology Explorer — none of which were yet catalogued; all three were already on the deferred queue. Three new entries this run, at the 5-entry soft cap minus two; Cortellis / Medidata / Consensus standalone plugin entries kept on the deferred queue pending direct install-path confirmation.

Added

ChEMBL Connector (Categories: Chemistry, Drug Repurposing and Discovery) — Anthropic-packaged plugin and Claude.ai connector over EMBL-EBI’s ChEMBL bioactive-compound database; six tool calls covering compound search, target search, bioactivity, and mechanism-of-action (Anthropic tutorial, marketplace).
ToolUniverse (Categories: All) — Zitnik Lab / MIT Lincoln Laboratory MCP server bundling 600+ vetted scientific tools across literature, chemistry, omics, clinical trials, and knowledge graphs; installable directly via uvx tooluniverse or as a life-sciences marketplace plugin (Claude Code setup guide, PyPI, arXiv:2509.23426).
Owkin Pathology Explorer Connector (Categories: Drug Repurposing and Discovery, Translational Medicine) — remote MCP connector exposing Owkin’s H&E whole-slide image analysis agent for cell-type detection, tumour-microenvironment profiling, and cohort-level survival analysis on TCGA and partner cohorts (Anthropic tutorial, Owkin press release).

Updated

catalog/curator-state.md — Recently surfaced refreshed with this run’s three Drug Repurposing additions; Deferred queue gained Augmented-Nature ChEMBL-MCP-Server (community alternative now superseded for marketplace users by the first-party ChEMBL connector) and broke the previously-bundled “Standalone anthropics/life-sciences plugins” line into per-plugin Deferred entries (Cortellis, Medidata, Consensus).

Verified (no changes)

Existing Drug Repurposing entries (open-targets, drugbank, pubchem, pdb, alphafold, depmap, bio-research) spot-checked against upstream — install paths and pricing claims still valid; no field drift.
Manifest sweep of anthropics/life-sciences re-confirmed: the four previously-deferred plugins beyond ChEMBL / ToolUniverse / Owkin (Cortellis, Medidata, Consensus, plus bio-research-already-catalogued bundles) carry forward to the deferred queue.

2026-05-23

Directed pass on Translational Medicine (Saturday focus). Manifest sweep of anthropics/healthcare/.claude-plugin/marketplace.json (via web search of the release notes and README, since direct fetch is unavailable from the runner) surfaced three new Anthropic-published healthcare components beyond the two already catalogued (fhir-developer, prior-auth-review). A focused web search for ClinicalTrials.gov MCP servers identified cyanheads/clinicaltrialsgov-mcp-server as the highest-quality standalone option, with a hosted public instance and Apache-2.0 license. Four new entries this run — at the 5-entry soft cap minus one — with NPI Registry deferred to next run.

Added

clinical-trial-protocol (Anthropic Healthcare Plugin) (Categories: Translational Medicine, Drug Repurposing and Discovery) — Anthropic skill that drafts FDA/NIH-compliant Phase 2/3 clinical-trial protocols via a four-step waypoint workflow (regulatory classification → ClinicalTrials.gov landscape → sample-size calculation → protocol drafting) (source, tutorial).
CMS Coverage MCP (Anthropic Healthcare) (Categories: Translational Medicine) — Anthropic-published MCP over the CMS Local and National Coverage Determinations for Medicare prior-auth, appeals, and policy lookup (source, announcement).
ICD-10 Codes MCP (Anthropic Healthcare) (Categories: Translational Medicine) — Anthropic-published MCP for ICD-10-CM and ICD-10-PCS diagnosis and procedure code lookup, sourced from CMS and CDC (source).
ClinicalTrials.gov MCP Server (cyanheads) (Categories: Translational Medicine, Drug Repurposing and Discovery) — Apache-2.0 MCP over the ClinicalTrials.gov v2 API with full study retrieval, outcomes extraction, and patient-to-trial matching; hosted public instance at clinicaltrials.caseyjhand.com/mcp (source, npm).

Updated

catalog/curator-state.md — Recently surfaced refreshed with this run’s four Translational Medicine additions; Deferred queue gained NPI Registry MCP (the fourth healthcare-marketplace MCP, held back to respect the entry cap) and AACT Clinical Trials MCP (SQL-over-PostgreSQL alternative to the ClinicalTrials.gov v2 API).

Verified (no changes)

Manifest sweep of anthropics/life-sciences/marketplace.json re-confirmed: no diff vs. 2026-05-22; existing Owkin / ChEMBL / Cortellis / ToolUniverse / Consensus / Medidata deferred items carry forward.
Existing Translational Medicine pages (fhir-developer, prior-auth-review, fhir-wso2) spot-checked against upstream — install paths still valid; no field drift.

2026-05-22

Directed pass on Neuroscience (Friday focus). Three K-Dense scientific skills added — each is a distinct SKILL.md directory installable via the K-Dense-AI/claude-scientific-skills plugin marketplace or by manual clone of K-Dense-AI/scientific-agent-skills. Manifest sweep of anthropics/life-sciences shows no new plugin entries since 2026-05-21.

Added

BIDS (Claude Skill) (Categories: Neuroscience) — K-Dense skill for the Brain Imaging Data Structure standard: dataset creation, DICOM-to-BIDS conversion (HeuDiConv / dcm2bids / BIDScoin), PyBIDS BIDSLayout queries, validation against OpenNeuro / DANDI submission requirements, and BIDS-Apps integration (source).
Neuropixels-Analysis (Claude Skill) (Categories: Neuroscience) — K-Dense skill for end-to-end Neuropixels pipelines: SpikeGLX / Open Ephys / NWB loading, preprocessing, motion correction, and Kilosort4 / SpykingCircus2 / Mountainsort5 spike sorting via SpikeInterface (source).
NeuroKit2 (Claude Skill) (Categories: Neuroscience, Translational Medicine) — K-Dense skill wrapping NeuroKit2 for ECG / EEG / EDA / RSP / PPG / EMG / EOG analysis, HRV, and multi-modal psychophysiology (source).

Updated

catalog/curator-state.md — Recently surfaced list refreshed with this run’s three Neuroscience additions.

Verified (no changes)

Manifest sweep of anthropics/life-sciences re-confirmed: no new plugin entries since the 2026-05-21 sweep; deferred queue (Owkin, ChEMBL, Cortellis, ToolUniverse, Consensus, Medidata standalone plugins) retained for next run.

2026-05-21

Directed pass on Molecular and Cellular Biology (Thursday focus). Four K-Dense scientific skills added — each is a distinct SKILL.md directory installable via the K-Dense-AI/claude-scientific-skills plugin marketplace or by manual clone of K-Dense-AI/scientific-agent-skills. Also created catalog/curator-state.md per the new schema.

Added

AnnData (Claude Skill) (Categories: Immunology and Microbiology, Molecular and Cellular Biology, Neuroscience) — K-Dense skill teaching the AnnData annotated-data-matrix object model that underpins Scanpy and scvi-tools (source).
PyDESeq2 (Claude Skill) (Categories: Drug Repurposing and Discovery, Immunology and Microbiology, Molecular and Cellular Biology, Translational Medicine) — K-Dense skill for bulk RNA-seq differential expression via the Python reimplementation of DESeq2 (source).
gget (Claude Skill) (Categories: Drug Repurposing and Discovery, Immunology and Microbiology, Integrative Structural and Computational Biology, Molecular and Cellular Biology) — K-Dense skill wrapping the gget unified query API over Ensembl, UniProt, NCBI, PDB, COSMIC, and 15+ other genomics databases (source).
DepMap (Claude Skill) (Categories: Drug Repurposing and Discovery, Molecular and Cellular Biology, Translational Medicine) — K-Dense skill for Cancer Dependency Map CRISPR Chronos gene-effect, PRISM drug sensitivity, and OMICS data across hundreds of cancer cell lines (source).

Updated

catalog/curator-state.md — created per the schema in AGENT.md; carries forward allenbrain-mcp flag and the prior Deferred queue, with Cellxgene Census / scVelo / Arboreto added as next-run Molecular & Cellular Biology candidates.

Verified (no changes)

Manifest sweep of anthropics/life-sciences/marketplace.json re-confirmed: existing entries (pubmed, biorender, synapse, wiley-scholar-gateway, 10x-genomics, single-cell-rna-qc, instrument-data-to-allotrope, nextflow-development, scvi-tools, scientific-problem-selection, open-targets, adisinsight) all still present — no diff against the previously catalogued set.

2026-05-20 (topic-focused seeding pass)

A one-time directed sweep across all seven categories, seeded by per-category seed queries newly added to AGENT.md. Twenty new tool pages, organized by the day-of-week the rotating focus is now scheduled to revisit each category.

Added — Chemistry (Mondays)

RDKit Cheminformatics Skill (Categories: Chemistry) — K-Dense Claude skill providing RDKit recipes for SMILES parsing, descriptors, fingerprints, substructure search, reactions, and 2D/3D molecular generation (source).
PubChem MCP Server (Categories: Chemistry, Drug Repurposing and Discovery) — community MCP wrapping the PubChem compound database; complements bioactivity-focused ChEMBL (source).
RDKit MCP Server (TandemAI) (Categories: Chemistry, Drug Repurposing and Discovery) — full RDKit 2025.3.1 surface exposed as discrete MCP tools, for environments without local Python execution (source).

Added — Immunology and Microbiology (Tuesdays)

BioContextAI Knowledgebase MCP (Categories: All) — read-only MCP unifying 14+ biomedical databases (Antibody Registry, UniProt, STRING, AlphaFold, KEGG, Open Targets, …) (source).
scikit-bio (Claude Skill) (Categories: Immunology and Microbiology, Molecular and Cellular Biology) — K-Dense skill for microbiome ecology, alpha/beta diversity, ordination, PERMANOVA, phylogenetics (source).
FlowIO (Claude Skill) (Categories: Immunology and Microbiology) — K-Dense skill parsing Flow Cytometry Standard (FCS v2–3.1) files for immunophenotyping pipelines (source).

Added — Integrative Structural and Computational Biology (Wednesdays)

PDB MCP Server (Categories: Integrative Structural and Computational Biology, Drug Repurposing and Discovery) — Augmented Nature MCP wrapping the RCSB Protein Data Bank with UniProt cross-referencing and structure-quality lookups (source).
AlphaFold MCP Server (Categories: Immunology and Microbiology, Integrative Structural and Computational Biology, Molecular and Cellular Biology, Drug Repurposing and Discovery) — Augmented Nature MCP over the EBI AlphaFold Protein Structure Database; ~25 tools including pLDDT analysis and PyMOL/ChimeraX export (source).
Molecule-MCP (Categories: Chemistry, Integrative Structural and Computational Biology, Drug Repurposing and Discovery) — chatmol three-server bundle driving PyMOL, ChimeraX, and GROMACS MD simulations via natural language (source).

Added — Molecular and Cellular Biology (Thursdays)

Scanpy-MCP (Categories: All) — MCP wrapping the full Scanpy workflow (IO, QC, normalization, PCA, clustering, DE, plotting) for natural-language single-cell analysis (source).
UniProt MCP Server (Categories: Immunology and Microbiology, Integrative Structural and Computational Biology, Molecular and Cellular Biology, Drug Repurposing and Discovery) — Augmented Nature MCP with 26 tools over UniProt REST: domains, orthologs, PTMs, pathways, multi-format export (source).

Added — Neuroscience (Fridays)

Neurosift Tools MCP (Categories: Neuroscience) — Flatiron Institute MCP for DANDI / OpenNeuro discovery, NWB introspection, and PyNWB docs semantic search (source).
allenbrain-mcp (Categories: Neuroscience) — Alpha community wrapper exposing Allen Brain Atlas RMA queries, cell types, mouse connectivity, ontologies, and image / grid downloads (source). License unset upstream — flag for review before redistribution.
AIND Data MCP (Categories: Neuroscience) — official Allen Institute for Neural Dynamics MCP for V2 metadata DocDB queries and NWB introspection (source).

Added — Translational Medicine (Saturdays)

WSO2 FHIR MCP Server (Categories: Translational Medicine) — Apache-2.0 MCP for FHIR R4 CRUD against any EHR or sandbox FHIR API with SMART-on-FHIR auth (source).
fhir-developer (Anthropic Healthcare Plugin) (Categories: Translational Medicine) — Anthropic Claude Code plugin for authoring FHIR R4 resources with LOINC, SNOMED, and RxNorm validation (source).
prior-auth-review (Anthropic Healthcare Plugin) (Categories: Translational Medicine) — Anthropic plugin reviewing prior-authorization request documents against payer rules and surfacing gaps (source).

Added — Drug Repurposing and Discovery (Sundays)

Open Targets Plugin (Categories: Drug Repurposing and Discovery, Molecular and Cellular Biology, Translational Medicine) — Anthropic-packaged plugin wrapping the official Open Targets MCP for target-disease associations and target-prioritisation (source).
AdisInsight Plugin (Categories: Drug Repurposing and Discovery, Translational Medicine) — Springer Nature commercial plugin for drug-development pipeline, clinical-trial, and deal intelligence (source).
DrugBank MCP Server (Categories: Drug Repurposing and Discovery, Chemistry, Translational Medicine) — community MCP over a local DrugBank SQLite (17k+ drugs) with 16 query methods; requires user-supplied DrugBank XML license (source).

Updated

AGENT.md — added a Topic-focused rotation section with per-category seed queries and source pointers, plus a directed-pass procedure that runs alongside the existing manifest sweep on every daily run.
.github/workflows/curate.yml — derives today’s focus category from the UTC weekday and injects focus_category: into the run prompt.
All existing tool pages — dropped explicit nav_order so the new and existing pages sort alphabetically together in the sidebar.

Deferred — next-run priority

Catalog candidates from the directed-pass that warrant a follow-up entry but were not added in this seeding pass:

Standalone anthropics/life-sciences plugins beyond Open Targets and AdisInsight: owkin, chembl, cortellis, tooluniverse, consensus, medidata. Each is a discrete /plugin install <name>@life-sciences target and deserves its own entry distinct from the bundled bio-research umbrella.
OpenClaw Medical Skills Library — 869-skill MIT-licensed collection. Per the catalog rule, treat each individual skill (clinical-trial-design, pharmacovigilance, FHIR-developer, adverse-event detection, PyHealth) as a separate entry rather than the umbrella repo.
Ensembl MCP servers (munch-group/ensembl-mcp, effieklimi/ensembl-mcp-server) — both early-stage; revisit when one of them stabilises.
UCSC Genome MCP (hlydecker/ucsc-genome-mcp) — 12 tools over the UCSC Genome Browser API.
NCBI Datasets MCP (Augmented-Nature/NCBI-Datasets-MCP-Server) — 31 tools over the NCBI Datasets API.
OpenFDA MCP (Augmented-Nature/OpenFDA-MCP-Server, ythalorossy/openfda) — standalone OpenFDA wrappers; BioMCP already covers OpenFDA but the distinct install path may warrant a separate entry.
Azure FHIR MCP (erikhoward/azure-fhir-mcp-server) — Azure Health Data Services-specific FHIR adapter.

2026-05-20

Surface knowledge-work-plugins bio-research umbrella plugin

Cleared the top Deferred item — bio-research@anthropics/knowledge-work-plugins — using the plugin’s GitHub README, the parent knowledge-work-plugins repo, and the DeepWiki marketplace breakdown as primary sources. It is a pan-life-sciences plugin (literature, single-cell, sequencing, drug discovery, strategy), so tagged Categories: All and added to every category index. Also dropped benchling@life-sciences from the Deferred list: it was removed from the anthropics/life-sciences marketplace because the plugin system does not support tenant-specific URLs (per the Benchling DeepWiki page); Benchling remains reachable indirectly via the new bio-research umbrella.

Added

bio-research (Claude Code Plugin) (Categories: All) — Anthropic open-source umbrella plugin in anthropics/knowledge-work-plugins bundling 5 analysis skills (Literature Review, Single-Cell Analysis, Sequencing Pipeline, Drug Discovery, Research Strategy) and ~10 MCP connectors (PubMed, BioRender, bioRxiv, ClinicalTrials.gov, ChEMBL, Synapse, Wiley Scholar Gateway, Owkin, Open Targets, Benchling); install via /plugin marketplace add anthropics/knowledge-work-plugins then /plugin install bio-research@knowledge-work-plugins, then /start (plugin README, DeepWiki).

Updated

[catalog/entries.md] — New bio-research entry inserted alphabetically between Anthropic PubMed Connector and BioMCP; Recently surfaced refreshed (5-item rolling window: dropped nextflow-development, added bio-research at top); Deferred reshaped to flag individual knowledge-work-plugins connectors (bioRxiv, ChEMBL, ClinicalTrials.gov, Owkin, Open Targets) and the K-Dense-AI scientific-skills collection as next-run targets.
[catalog/README.md] — Distinct-tool count 11 → 12; per-category counts incremented by 1 across all seven indices (Chemistry 7 → 8; Structural/Computational and Neuroscience 10 → 11; Immunology, Molecular & Cellular, Drug Discovery, Translational 11 → 12). All-tagged-tool list updated to include bio-research.
[All seven category indices] — One new bio-research (Claude Code Plugin) card added per index in the alphabetical slot between Anthropic PubMed Connector and BioMCP.
[Deferred list] — Removed bio-research@anthropics/knowledge-work-plugins (surfaced this run); recorded benchling@life-sciences as dropped with rationale and DeepWiki citation; added individual bundled connectors and K-Dense-AI/scientific-agent-skills as next-run priorities.

Flagged

None.

Verified (no changes)

11 existing entries spot-checked (all carry Last verified between 2026-05-19 and 2026-05-20 from recent runs); no field drift on supplier links, availability, or install paths. The bio-research plugin’s bundled PubMed/BioRender/Synapse/Wiley connectors corroborate the existing standalone entries for those tools.

2026-05-20

Surface anthropics/life-sciences marketplace entries (batch 4)

Cleared all three remaining *@life-sciences candidates from the Deferred list — Synapse, Wiley Scholar Gateway, and scientific-problem-selection — using web search results that cite the anthropics/life-sciences marketplace, the Anthropic tutorials, and the DeepWiki marketplace breakdown as primary sources. All three are pan-life-sciences tools and tagged Categories: All, so each was added to every category index. Held at three new entries this run (within the soft cap of 5) since each All-tagged entry costs 1 entry-block write + 7 card edits = 8 file touches.

Added

Scholar Gateway Connector (Wiley) (Categories: All) — Wiley remote MCP server / Claude.ai connector providing peer-reviewed scholarly content across 3M+ articles (including 300+ Life Sciences journals covering 900,000+ research articles); hosted at connector.scholargateway.ai/mcp, Beta, free Scholar Gateway account required (Anthropic tutorial, DeepWiki, Wiley press release).
scientific-problem-selection (Claude Skill) (Categories: All) — Anthropic skill encoding Fischbach & Walsh’s Cell (2024) framework for project ideation, risk assessment, troubleshooting stuck projects, and strategic scientific planning (marketplace, Claude for Life Sciences).
Synapse.org Connector (Categories: All) — Sage Bionetworks remote MCP server / Claude.ai connector at mcp.synapse.org/mcp for discovery, project structure, and metadata retrieval across Synapse-hosted biomedical data; OAuth2 default, free Synapse account required, per-project access controls apply (Anthropic tutorial, marketplace, server source).

Updated

[catalog/entries.md] — Three new entry blocks inserted alphabetically; Recently surfaced refreshed (5-item rolling window: dropped instrument-data-to-allotrope, BioRender Connector, 10x Genomics Cloud MCP; added Synapse, scientific-problem-selection, Scholar Gateway at top).
[catalog/README.md] — Distinct-tool count 8 → 11; per-category counts incremented by 3 across all seven indices (Chemistry 4 → 7; Structural/Computational and Neuroscience 7 → 10; Immunology, Molecular & Cellular, Drug Discovery, Translational 8 → 11). All-tagged-tool list updated to include Scholar Gateway, Synapse, and scientific-problem-selection.
[All seven category indices] — Three new cards added per index in their alphabetical slots, freshness timestamps bumped.
[Deferred list] — Removed all three surfaced items; bio-research@anthropics/knowledge-work-plugins carries forward; added benchling@life-sciences (named in Anthropic launch coverage alongside 10x Genomics) as a new next-run priority.

Flagged

None.

Verified (no changes)

8 existing entries spot-checked (all Last verified: 2026-05-19 from prior runs); no field drift on supplier links, availability, or install paths.

2026-05-19

Surface anthropics/life-sciences marketplace entries (batch 3)

Cleared one more item from the Deferred list — scvi-tools@life-sciences — using the anthropics/life-sciences marketplace plus the Anthropic tutorial as primary sources. Held to one new entry this run because the tag list expands to six category-card edits (Chemistry excluded — scvi-tools is single-cell omics, not chemistry).

Added

scvi-tools (Claude Skill) (Categories: Drug Repurposing and Discovery, Immunology and Microbiology, Integrative Structural and Computational Biology, Molecular and Cellular Biology, Neuroscience, Translational Medicine) — Anthropic skill bundling deep-learning workflows for scVI, scANVI, totalVI, MultiVI, PeakVI, DestVI, contrastiveVI, sysVI, and veloVI; covers batch correction, semi-supervised cell-type annotation, multi-modal CITE-seq / multiome integration, spatial deconvolution, perturbation analysis, and RNA velocity (Anthropic tutorial, marketplace).

Updated

[catalog/entries.md] — Promoted scvi-tools from the Deferred list; refreshed Recently surfaced (5-item rolling window: dropped single-cell-rna-qc, added scvi-tools at top).
[catalog/README.md] — Distinct-tool count 7 → 8; per-category counts incremented for the six tagged categories (Immunology, Molecular & Cellular, Drug Discovery, Translational 7 → 8; Structural/Computational and Neuroscience 6 → 7). Chemistry unchanged at 4.
[Deferred list] — Removed scvi-tools@life-sciences; remaining items (synapse, wiley-scholar-gateway, scientific-problem-selection, bio-research umbrella) carry forward.

Flagged

None.

Verified (no changes)

7 existing entries spot-checked (all Last verified: 2026-05-19 from prior runs today); no field drift on supplier links or install paths.

2026-05-19

Surface anthropics/life-sciences marketplace entries (batch 2)

Cleared two items from the previous run’s Deferred list using the anthropics/life-sciences marketplace as the manifest-driven source. Stayed under the per-run surfacing cap (2 entries this run vs. soft cap of 5) because each new entry’s tag list expands into 5–6 category-card edits.

Added

instrument-data-to-allotrope (Claude Skill) (Categories: Chemistry, Drug Repurposing and Discovery, Immunology and Microbiology, Molecular and Cellular Biology, Translational Medicine) — Anthropic skill that converts 40+ lab-instrument output formats (PDF, CSV, Excel, TXT) to Allotrope Simple Model JSON / flattened CSV via the allotropy library, with PDF-table fallback and ASM validation (skill listing, marketplace).
nextflow-development (Claude Skill) (Categories: Drug Repurposing and Discovery, Immunology and Microbiology, Integrative Structural and Computational Biology, Molecular and Cellular Biology, Neuroscience, Translational Medicine) — Anthropic prototype skill that runs nf-core rnaseq 3.22.2, sarek 3.7.1, and atacseq 2.1.2 on local FASTQ or GEO/SRA inputs (skill listing, marketplace).

Updated

[catalog/entries.md] — Promoted both skills from the Deferred list; refreshed Recently surfaced to include them.
[catalog/README.md] — Distinct-tool count 5 → 7; per-category counts incremented (Chemistry 3 → 4; Structural/Computational and Neuroscience 5 → 6; Immunology, Molecular & Cellular, Drug Discovery, and Translational 5 → 7).
[Deferred list] — Removed the two surfaced items; added bio-research@anthropics/knowledge-work-plugins (umbrella plugin bundling 10 MCP servers + 6 skills) as a new next-run priority after it was named by both search results.

Flagged

None.

Verified (no changes)

5 existing entries spot-checked (all Last verified: 2026-05-19 from yesterday’s run); no field drift on supplier links or install paths.

2026-05-19

Refactor: entries.md is canonical; category files become tag-filtered indices

Multi-category entries were previously duplicated as full blocks into every named category file. That hit the action’s 600-second wall on expansion runs (each pan-life-sciences entry cost ~7 file writes plus drift detection) and forced an artificial “primary category” choice for genuinely cross-cutting tools like PubMed.

Refactored to a single source of truth:

catalog/entries.md now holds every entry’s full content block, alphabetically.
Each catalog/<area>.md is a card-based index: brief summary per entry tagged with that category, linking back to entries.md.
The Categories field is a tag list (comma-separated), with the literal value All allowed for tools applicable across every life-science domain. No notion of a “primary” category.
Drift detection at run-start is no longer needed — single source of truth, no drift possible.
Updates to a tool’s pricing/availability/etc. are one write to entries.md. Card edits in indices are only needed when the card’s surface text (name, type, supplier, availability, one-line summary) changes.

This refactor was performed locally rather than by the agent because it’s a substantial content migration and the agent’s per-run timeouts made it unsafe to run there. AGENT.md was rewritten in the same change to match the new storage model.

Removed

Per-category duplicated copies of the 5 cross-cutting entries (Anthropic PubMed Connector, BioMCP, BioRender Connector, 10x Genomics Cloud MCP, single-cell-rna-qc) — superseded by entries.md as canonical store.

Added

catalog/entries.md — canonical file with all 5 entries; PubMed / BioMCP / BioRender tagged All, 10x Genomics Cloud / single-cell-rna-qc tagged with their explicit 6-category subset.
Card-based index in each category file linking back to canonical entries.

2026-05-19

Surface anthropics/life-sciences marketplace entries (batch 1)

Drew from the anthropics/life-sciences marketplace manifest — the highest-yield, pre-validated source for Claude-installable life-science components. Added the three most impactful entries not yet catalogued and stopped at the per-run cap of 3.

Added

[Chemistry, Immunology and Microbiology, Integrative Structural and Computational Biology, Molecular and Cellular Biology, Neuroscience, Translational Medicine, Drug Repurposing and Discovery] BioRender Connector — Scientific-illustration MCP / Claude.ai connector launched alongside the Oct 23, 2025 Anthropic × BioRender partnership; relevant across every life-science domain because figure-building is universal (Anthropic tutorial, BusinessWire announcement).
[Immunology and Microbiology, Integrative Structural and Computational Biology, Molecular and Cellular Biology, Neuroscience, Translational Medicine, Drug Repurposing and Discovery] 10x Genomics Cloud MCP — Local MCPB extension distributed by 10x Genomics that lets Claude Code and Claude Desktop drive 10x Cloud single-cell / immune-profiling / Visium / Xenium analyses; available from Oct 20, 2025 (10x docs).
[Immunology and Microbiology, Integrative Structural and Computational Biology, Molecular and Cellular Biology, Neuroscience, Translational Medicine, Drug Repurposing and Discovery] single-cell-rna-qc (Claude Skill) — Anthropic’s first published scientific skill; performs scverse MAD-based QC on .h5ad and 10x .h5 inputs (SKILL.md, Anthropic tutorial).

Updated

[catalog/README.md] — Refreshed entry counts (Chemistry → 3, all other categories → 5; distinct tools → 5) and updated cross-cutting-tools example list.

Flagged

None.

Verified (no changes)

None — new-surfacing run.

Deferred — next-run priority

The following candidates were observed in anthropics/life-sciences and adjacent sources during this run but deferred to remain within the 3-entry surfacing cap. Pick up next run before re-querying any source.

synapse@life-sciences — Remote MCP server for Sage Bionetworks Synapse; needs verification of auth model and supplier-side docs.
wiley-scholar-gateway@life-sciences — Remote MCP server for Wiley scholarly content; verify subscription/auth requirements.
instrument-data-to-allotrope@life-sciences — Skill that converts lab-instrument data to Allotrope Simple Model; useful in Chemistry and Translational Medicine.
nextflow-development@life-sciences — Skill for running nf-core pipelines (rnaseq, sarek, atacseq) on local or GEO/SRA inputs.
scvi-tools@life-sciences — Skill packaging the scvi-tools deep-learning toolkit for single-cell omics.
scientific-problem-selection@life-sciences — Skill encoding Fischbach & Walsh (Cell 2024) scientific-project framework.

2026-05-19

Enable multi-category entries

Backfilled the new Categories field on every existing catalog entry and duplicated each entry block byte-identically into every category file it claims. Both currently-catalogued tools are cross-cutting biomedical research infrastructure and were assigned all seven categories.

Updated

[All categories] Anthropic PubMed Connector — Added Categories: Chemistry | Immunology and Microbiology | Integrative Structural and Computational Biology | Molecular and Cellular Biology | Neuroscience | Translational Medicine | Drug Repurposing and Discovery. Entry block duplicated into the six non-Translational-Medicine category files.
[All categories] BioMCP — Added Categories: Chemistry | Immunology and Microbiology | Integrative Structural and Computational Biology | Molecular and Cellular Biology | Neuroscience | Translational Medicine | Drug Repurposing and Discovery (reflecting that it bundles ClinicalTrials.gov, PubMed, MyVariant.info, and OpenFDA — all relevant to every life-science domain). Entry block duplicated into the six non-Translational-Medicine category files.
[catalog/README.md] — Refreshed entry counts (each category now lists 2 entries; 2 distinct tools across the catalog) and clarified that cross-cutting entries are duplicated across files.
[AGENT.md] — Removed the one-time multi-category backfill subsection now that the backfill has been executed.

Added

[Chemistry] Anthropic PubMed Connector — Cross-cutting literature search; added to this category’s Recently surfaced (Tutorial).
[Chemistry] BioMCP — Cross-cutting biomedical MCP; added to this category’s Recently surfaced (biomcp.org).
[Immunology and Microbiology] Anthropic PubMed Connector — Cross-cutting literature search; added to this category’s Recently surfaced.
[Immunology and Microbiology] BioMCP — Cross-cutting biomedical MCP; added to this category’s Recently surfaced.
[Integrative Structural and Computational Biology] Anthropic PubMed Connector — Cross-cutting literature search; added to this category’s Recently surfaced.
[Integrative Structural and Computational Biology] BioMCP — Cross-cutting biomedical MCP; added to this category’s Recently surfaced.
[Molecular and Cellular Biology] Anthropic PubMed Connector — Cross-cutting literature search; added to this category’s Recently surfaced.
[Molecular and Cellular Biology] BioMCP — Cross-cutting biomedical MCP; added to this category’s Recently surfaced.
[Neuroscience] Anthropic PubMed Connector — Cross-cutting literature search; added to this category’s Recently surfaced.
[Neuroscience] BioMCP — Cross-cutting biomedical MCP; added to this category’s Recently surfaced.
[Drug Repurposing and Discovery] Anthropic PubMed Connector — Cross-cutting literature search; added to this category’s Recently surfaced.
[Drug Repurposing and Discovery] BioMCP — Cross-cutting biomedical MCP; added to this category’s Recently surfaced.

Flagged

None.

Verified (no changes)

None — backfill only.

2026-05-19

Scope refocus to Claude-installable components

The catalog scope was narrowed to discrete, installable Claude components (Claude Skills, MCP servers, Claude Code Plugins, Claude.ai Connectors). General-purpose libraries, model weights distributed only as research artifacts, hosted SaaS without a Claude-installable wrapper, and bespoke LangChain-style agents are no longer in scope. The entry schema was migrated to add Available in (every supported install path) and Tools / resources exposed, and to drop the free-form Benchmarks and Installation fields. Surviving entries were re-keyed to the new schema; out-of-scope entries were removed (git history preserves them).

Removed

[Chemistry] RDKit — General-purpose cheminformatics library; no Claude Skill/MCP/Plugin/Connector wrapper.
[Chemistry] DeepChem — General-purpose ML toolkit for chemistry; no Claude-installable wrapper.
[Chemistry] ChemCrow — LangChain bespoke agent, not packaged as a Skill or Plugin.
[Structural and Computational Biology] AlphaFold 3 — Model weights + hosted server; no Claude-installable wrapper.
[Structural and Computational Biology] Boltz-1 / Boltz-2 — Model weights / research code; no Claude-installable wrapper.
[Structural and Computational Biology] Chai-1 — Model weights + hosted SaaS; no Claude-installable wrapper.
[Structural and Computational Biology] OpenFold / OpenFold3 — Research code reproductions; no Claude-installable wrapper.
[Structural and Computational Biology] ESM-2 / ESMFold — Model weights distributed as research artifacts; no Claude-installable wrapper.
[Structural and Computational Biology] RFdiffusion — Research-artifact model weights; no Claude-installable wrapper.
[Immunology and Microbiology] IgFold — Research-artifact library; no Claude-installable wrapper.
[Molecular and Cellular Biology] Scanpy — General-purpose single-cell library; no Claude-installable wrapper.
[Molecular and Cellular Biology] CZ CELLxGENE Discover Census — Hosted dataset + client library; no Claude-installable wrapper.
[Neuroscience] DeepLabCut — General-purpose pose-estimation library; no Claude-installable wrapper.
[Translational Medicine] Claude for Life Sciences — Umbrella offering, not a discrete installable component.
[Drug Repurposing and Discovery] TxGNN — Research-artifact model and code; no Claude-installable wrapper.

Updated

[Translational Medicine] Anthropic PubMed Connector — Re-keyed to new schema with explicit Available in (Claude Code plugin marketplace, direct mcp add, Claude.ai Healthcare connector) and Tools / resources exposed.
[Translational Medicine] BioMCP — Re-keyed to new schema with explicit Available in (Claude Code via uv, Claude Desktop mcp_config.json) and Tools / resources exposed.
[AGENT.md] — Removed the one-time scope-migration subsection now that it has been executed.
[catalog/README.md] — Refreshed entry counts and timestamp; updated schema summary.

Flagged

None.

Verified (no changes)

None — scope migration only.

2026-05-18

First substantive curator run — seeded each category with established, primary-source-verifiable entries.

Added

[Chemistry] RDKit — Core BSD-3-Clause cheminformatics toolkit; latest release 2025.09.x (GitHub, Install docs)
[Chemistry] DeepChem — MIT-licensed deep-learning toolkit for chemistry and materials (GitHub)
[Chemistry] ChemCrow — LangChain chemistry agent (Bran et al., Nature Machine Intelligence 2024) (GitHub)
[Structural and Computational Biology] AlphaFold 3 — DeepMind biomolecular structure prediction; server GA, code under CC-BY-NC-SA 4.0 (GitHub)
[Structural and Computational Biology] Boltz-1 / Boltz-2 — MIT-licensed open structure + affinity models (MIT Jameel Clinic) (GitHub)
[Structural and Computational Biology] Chai-1 — Chai Discovery multimodal predictor with free web interface (GitHub)
[Structural and Computational Biology] OpenFold / OpenFold3 — Apache-2.0 reproductions of AlphaFold 2/3 (GitHub, OpenFold3)
[Structural and Computational Biology] ESM-2 / ESMFold — Meta FAIR protein language model + folding head (GitHub)
[Structural and Computational Biology] RFdiffusion — Baker Lab generative protein design; RFdiffusion3 released Dec 2025 (GitHub, IPD announcement)
[Immunology and Microbiology] IgFold — Gray Lab antibody structure prediction with AntiBERTy embeddings (GitHub)
[Molecular and Cellular Biology] Scanpy — scverse single-cell analysis toolkit, BSD-3-Clause, v1.12.1 (GitHub)
[Molecular and Cellular Biology] CZ CELLxGENE Discover Census — CZI hosted single-cell corpus with Python/R APIs (Docs)
[Neuroscience] DeepLabCut — Mathis Labs markerless pose estimation; v3.0 PyTorch engine (GitHub)
[Translational Medicine] Anthropic PubMed Connector — Official MCP server for NCBI literature (Tutorial)
[Translational Medicine] BioMCP — GenomOncology MIT-licensed MCP server bundling ClinicalTrials.gov, PubMed, MyVariant, OpenFDA (biomcp.org)
[Translational Medicine] Claude for Life Sciences — Anthropic life-science offering launched Oct 2025 (CNBC)
[Drug Repurposing and Discovery] TxGNN — Zitnik Lab zero-shot graph foundation model for drug repurposing (Huang et al., Nature Medicine 2024) (GitHub)

Updated

[catalog/README.md] — Refreshed with current entry counts and freshness timestamp.

Flagged

None.

Verified (no changes)

None — first substantive run.