SpikeLab

A suite of computational-neuroscience agent skills for loading, sorting, analyzing, and visualizing neuronal spike-train data from multi-electrode-array (MEA) electrophysiology experiments.


Type	Claude Skill (bundled with an optional MCP server)
Supplier	Braingeneers
Availability	Beta — bioRxiv preprint posted 2026-04-25
Pricing	Free / OSS (MIT)
Capabilities	Read/Write — Claude runs the skills’ Python locally (Bash); the sorter writes results, the educator skill is read-only

How to install

SpikeLab ships its skills inside the installed Python package at spikelab/agent/skills/. A lightweight router skill detects the environment and delegates to the specialized skills.

Install the package (with the MCP extra if you also want the MCP server):
```
pip install "spikelab[all]"
```
(Add the mcp extra explicitly if you used a narrower install — e.g. pip install "spikelab[s3,ml,mcp]". Kilosort4 is not pulled in by [all]; install it separately if you need that sorter.)
Make the skills visible to Claude Code — copy or symlink them into your skills directory:
```
cp -r "$(python -c 'import spikelab, os; print(os.path.join(os.path.dirname(spikelab.__file__), "agent", "skills"))')" ~/.claude/skills/spikelab
```
(The python -c … resolves the installed package’s agent/skills path for you; replace ~/.claude/skills/spikelab with your agent’s skills directory if different. CLI agents that auto-load skills from installed packages pick these up without the copy.)
(Optional) register the MCP server for programmatic tool access instead of skills. The upstream README documents the MCP extra but does not publish a copy-pasteable claude mcp add snippet — see the project README and adapt the stdio form below to the module/entry point it ships.
- Claude Code:
```
claude mcp add --transport stdio spikelab -- python -m spikelab.agent.mcp
```
- Claude Desktop (claude_desktop_config.json):
```
{
  "mcpServers": {
    "spikelab": { "command": "python", "args": ["-m", "spikelab.agent.mcp"] }
  }
}
```
  (Both snippets assume the MCP entry point is spikelab.agent.mcp; Unverified — confirm the exact module/console-script from the README before relying on it. The MCP server is launched by Claude over stdio — no long-running process to keep open.)

What it does

Skills bundled in the package:

spikelab-spikesorter — sort raw recordings with Kilosort2, Kilosort4, or RT-Sort; curate units; run stim-aligned sorting; inspect sorting output (can submit compute-heavy jobs to remote Kubernetes clusters).
spikelab-analysis-implementer — load data, write and run analysis scripts, generate figures, manage results.
spikelab-educator — explain analyses and methods (read-only).
spikelab-developer — promote analysis code into the library, write tests, expose methods via MCP, open PRs.
spikelab-map-updater — regenerate repo map files after changes.

Loads HDF5, NWB, KiloSort/Phy, and SpikeInterface formats; represents spike trains as SpikeData objects (per-unit spike times in ms) and firing rates as RateData.

Primary use cases: spike sorting MEA recordings, spike-train analysis and figure generation, teaching electrophysiology methods.

Notes

Python package; the spike sorters (Kilosort2/4, RT-Sort) have heavy GPU/compute dependencies and the package can offload to remote Kubernetes. The MCP server is optional and only present with the mcp extra. For a format-agnostic multi-sorter skill without the MEA/Kubernetes focus, see SpikeInterface; for Neuropixels-specific workflows see Neuropixels Analysis.

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.