MNE-MCP — Claude Code MCP server: EEG, MEG, fNIRS analysis
MCP server for neurophysiology data analysis (EEG, MEG, sEEG, ECoG, fNIRS) through MNE-Python. Describe the task — the agent loads the recording, filters it, builds epochs, computes ERPs, and saves plots.
claude mcp add mne-mcp -- uvx mne-mcpWhat MNE-MCP does
The server holds one MNE-Python session: load a Raw object once, then filter, re-reference, run ICA, epoch, average — all in memory, no reloading. Each step saves a PNG (power spectrum, sensor maps, ICA components, ERPs); the agent reads them and explains results. Comes with 41 tools for typical pipelines (filtering, ICA, epoching, averaging) plus work at source level (localization, connectivity, decoding), and an mne_run_code escape hatch for any MNE API call in the same live session.
Requires Python 3.12+, Git, and an MCP client. Line frequency, montage, filter band, rejection threshold, ICA parameters, epoch window, directories, and timeout are set through mne-mcp configure. Install via pip as mne-mcp, then mne-mcp setup registers your chosen client (Claude Code, Codex, PsyClaw, opencode) and installs 14 skills. MNE libraries and scientific dependencies are user-managed — the lightweight wrapper does not pull them.
After restarting the client, check status by calling mne_check_status: connected and mne_available are independent flags — the server connects even when MNE is missing.
What it helps with
- analyze EEG, MEG, sEEG, ECoG, and fNIRS recordings through a conversational interface
- filter signal, remove eye artifacts via ICA, and build epochs without reloading the file
- visualize power spectra, sensor maps, ICA components, and ERPs, saving PNGs for review
- run source localization, connectivity analysis, and decoding through structured tools
- execute arbitrary MNE-Python code in the live session via
mne_run_code
Example prompts for the agent
- Load
sub-01_raw.fifand show the power spectrum - Apply a 1–40 Hz bandpass filter and a 50 Hz notch filter, then run ICA to remove eye artifacts
- Build epochs around
stimulus/visualevents, average them, and save the ERP
How to install MNE-MCP
Platform CLI
claude mcp add mne-mcp -- uvx mne-mcpuvx runs the command named after the package; if the README names another one, use uvx --from mne-mcp <command>
Platform CLI
codex mcp add mne-mcp -- uvx mne-mcpuvx runs the command named after the package; if the README names another one, use uvx --from mne-mcp <command>
Config file
{
"mcpServers": {
"mne-mcp": {
"command": "uvx",
"args": [
"mne-mcp"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from mne-mcp <command>
Config file
{
"servers": {
"mne-mcp": {
"type": "stdio",
"command": "uvx",
"args": [
"mne-mcp"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from mne-mcp <command>
Platform CLI
gemini mcp add mne-mcp uvx mne-mcpuvx runs the command named after the package; if the README names another one, use uvx --from mne-mcp <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about MNE-MCP
How do I install MNE-MCP in Claude Code?
Activate a Python 3.12+ environment with MNE, run python -m pip install mne-mcp, then mne-mcp setup --clients claude. Restart Claude Code and check status by calling mne_check_status.
What does MNE-MCP do?
Gives the agent access to MNE-Python for neurophysiology data analysis. Keeps one session in memory: load a recording once, then filter, run ICA, build epochs, compute ERPs — all without reloading. Saves plots as PNGs; the agent reads and explains them.
Which libraries does MNE-MCP need?
Python 3.12+, Git, and MNE-Python with its scientific dependencies. The mne-mcp package is a lightweight wrapper and does not install them. Manage your environment yourself, then register the server with mne-mcp setup.
