neuro-mcp

MCP server that gives Claude an end-to-end EEG/MEG workflow — MNE-Python preprocessing, ICA, ERP and time-frequency analysis, source imaging, plus a BIDS-backed subject/EHR record store with an audit log.


Type	MCP server
Supplier	AImplifier
Availability	Alpha (PyPI 0.1.3, released 2026-07-24; MCP Registry `io.github.AImplifier/neuro-mcp`)
Pricing	Free / OSS — BSD-3-Clause
Capabilities	Read/Write — reads recordings, writes derived files, subject records, annotations, and EHR entries to a local database
Verified	works · 2026-08-17
Security	caution · 2026-08-17 — BSD-3-Clause and provenance confirmed; Alpha/single-org project writing local subject and EHR records

How to install

Requires Python 3.10+. A dedicated environment is recommended because MNE pulls a large scientific stack.

Create an environment and install (the upstream README uses conda):
```
conda create -n neuro-mcp python=3.11 -y
conda activate neuro-mcp
pip install neuro-mcp
```
Optional extras — Postgres backend and 3-D source visualization:
```
pip install "neuro-mcp[postgres]"
pip install "neuro-mcp[viz3d]"
```
(python -m venv + pip install neuro-mcp also works; conda is not required.)
Verify it starts (one-shot; Ctrl-C once it boots — Claude launches the process itself over stdio):
```
python -m neuro_mcp
```
Claude Code — direct MCP add (stdio). Use the interpreter inside the environment created in step 1, not a bare python:
```
claude mcp add --transport stdio neuro-analysis -- /path/to/envs/neuro-mcp/bin/python -m neuro_mcp
```
(replace /path/to/envs/neuro-mcp/bin/python with the absolute path printed by which python while the environment is active — e.g. /Users/you/miniconda3/envs/neuro-mcp/bin/python.)

Claude Desktop — add to claude_desktop_config.json:

{
  "mcpServers": {
    "neuro-analysis": {
      "command": "/path/to/envs/neuro-mcp/bin/python",
      "args": ["-m", "neuro_mcp"],
      "env": {
        "DATABASE_URL": "sqlite:////data/neuro_mcp.db",
        "BIDS_ROOT": "/data/bids"
      }
    }
  }
}

(same substitution for the interpreter path; adjust DATABASE_URL and BIDS_ROOT to writable locations on your machine.)

What it does

54 tools across three layers:

Signal processing (MNE-Python) — load_neuro, filter_neuro, resample_neuro, set_montage, set_reference, detect_bad_channels, run_ica, apply_ica, find_events, epoch_neuro, compute_psd, compute_erp, time_frequency, and a family of plot_* tools.
Source imaging (ESI) — template-head fetching through to extract_label_timecourses, i.e. forward model → inverse solution → parcellated label time courses.
Data and record store — subjects (register_subject, get_subject), versioned EHR entries (add_ehr_record, amend_ehr_record, void_ehr_record, get_ehr_history), datasets and recordings (import_recording, register_dataset, query_datasets, list_recordings), annotations (add_annotation, update_annotation, void_annotation, list_annotations), and get_audit_log.
NeuroII visualization — neuroii_push_recording, neuroii_create_viz_session, neuroii_pull_annotations, plus visualize_timeseries, visualize_averaging, visualize_esi. Views are exported as self-contained interactive Plotly HTML files that work offline; the hosted NeuroII service is optional and configured via NEUROII_API_URL / NEUROII_API_TOKEN.

Environment variables: DATABASE_URL (default sqlite:///~/.neuro-mcp/neuro_mcp.db), BIDS_ROOT (default ~/.neuro-mcp/bids), NEURO_MCP_HOME, NEUROII_API_URL, NEUROII_API_TOKEN.

Primary use cases: conversational EEG/MEG preprocessing and ERP analysis, EEG source localization, keeping subject records and annotations alongside BIDS recordings.

Notes

Not a cleared clinical device. The project describes itself as assisting clinicians and researchers, but publishes no regulatory clearance and no medical disclaimer. Treat every output as research-grade; do not use it for diagnosis or care decisions.
It writes. Unlike most catalogued neuroscience servers this one persists state — subject records, EHR entries, annotations, imported recordings. Point DATABASE_URL and BIDS_ROOT at scratch locations before letting an agent loose on real data, and note that the EHR tools amend/void rather than delete (there is a get_audit_log).
Defaults need no infrastructure: SQLite plus a scratch BIDS directory runs with zero setup; DATABASE_URL pointed at Postgres is the multi-user path.
3-D source rendering needs the viz3d extra; without it the ESI tools still compute but the 3-D views are unavailable.
Very early and unproven: 0 GitHub stars, version 0.1.3, single organization. There is no publication and no independent evaluation. Verify any preprocessing result against a hand-run MNE pipeline before relying on it.
Complements rather than replaces MNE-Python (EEG) (Claude Skill) — the skill teaches Claude to write MNE code, this server exposes MNE operations as callable tools. Related: EEG (Claude Skill), MEG (Claude Skill), BIDS, NeuroKit2.

Sources

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.