ToolUniverse
MCP server developed by the Zitnik Lab at Harvard Medical School with MIT Lincoln Laboratory, exposing 600+ vetted scientific tools spanning literature search, cheminformatics, omics, and clinical trial discovery. Designed as the tool backbone for AI-scientist workflows.
| Type | MCP server (also distributed as MCPB bundle and Claude.ai extension) |
| Supplier | Zitnik Lab, Harvard Medical School; listed in anthropics/life-sciences marketplace |
| Availability | GA — PyPI tooluniverse 1.0.x, surfaced as an Anthropic Life Sciences connector at J.P. Morgan 2026 |
| Pricing | Free / OSS; some wrapped APIs require user-supplied keys (e.g., Azure OpenAI for AI-tool-search features) |
| Capabilities | Read/Write — most tools are read-only over public APIs; a few orchestrate analysis steps |
How to install
- Claude Code — direct stdio via
uvx(simplest; no environment to manage):claude mcp add --transport stdio tooluniverse -- uvx tooluniverse - Claude Code —
anthropics/life-sciencesmarketplace (if you already use the marketplace):/plugin marketplace add anthropics/life-sciences /plugin install tooluniverse@life-sciences - Claude Desktop — MCPB bundle: download the
.mcpbbundle from the ToolUniverse releases page and double-click to install via the Claude Desktop extensions UI. - Manual
uvenvironment (useful if you want a filtered tool subset or to pin a version):# 1) install uv (https://docs.astral.sh/uv/) # 2) create and populate a dedicated env (replace /abs/path/to/claude_toolu_env with a directory you choose — e.g., $HOME/.tooluniverse-env) mkdir -p /abs/path/to/claude_toolu_env uv --directory /abs/path/to/claude_toolu_env pip install tooluniverse # 3) register with Claude Code claude mcp add tooluniverse --scope local -- uv --directory /abs/path/to/claude_toolu_env run tooluniverse-smcp-stdioFilter to a research-focused subset with
--include-tools EuropePMC_search_articles,ChEMBL_search_similar_molecules,openalex_literature_search,search_clinical_trialsappended to thetooluniverse-smcp-stdioinvocation.
What it does
A single MCP surface that exposes 600+ tools covering:
- Literature — EuropePMC, OpenAlex, PubMed search and full-text retrieval
- Chemistry / pharmacology — ChEMBL, PubChem, drug-target lookups, similarity search
- Omics & targets — Open Targets, Ensembl, UniProt, gene/protein cross-references
- Clinical trials & regulatory — ClinicalTrials.gov search, FDA datasets
- Knowledge graphs and ontologies — disease, phenotype, pathway lookups
Includes an AI-powered tool-search facility that lets Claude pick the right tool from the catalogue at runtime rather than overwhelming the context window with 600 definitions at once.
Primary use cases: AI-scientist agents that need broad coverage; hypothesis generation and iterative refinement across literature → omics → chemistry → trials; comparing competing hypotheses by pulling evidence from multiple databases in one session.
Notes
stdio transport when registered locally. Many wrapped APIs are public, but the AI-tool-search feature can be configured to use an LLM via environment variables (e.g., AZURE_OPENAI_API_KEY, AZURE_OPENAI_ENDPOINT). The MCPB bundle entry point is tooluniverse-smcp-stdio; the uvx tooluniverse shortcut is preferred for Claude Code because it doesn’t require maintaining a separate env.
Sources
- ToolUniverse — Claude Code setup guide
- Anthropic tutorial — Using the ToolUniverse Extension in Claude
tooluniverseon PyPI- Democratizing AI scientists using ToolUniverse (arXiv:2509.23426)
anthropics/life-sciencesmarketplace
Installed this tool?
Share feedback — install path, OS, errors, workarounds. The form opens with this tool pre-selected and a link back to this page.