# Nimbus BCI [Health: Active]

**Category:** Biology, Medicine and Bioinformatics  
**Repository:** https://github.com/nimbusbci/nimbus-mcp  
**GitHub Stars:** 3  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/nimbus-bci

## Description
AI agents build, train, and analyze BCI/EEG pipelines: data, models, experiments, live sessions.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "nimbus-bci": {
    "command": "uvx",
    "args": ["nimbus-mcp"],
    "env": {
      "NIMBUS_API_URL": "",
      "NIMBUS_TOKEN": "",
      "NIMBUS_TOKEN_FILE": "",
      "NIMBUS_MCP_KEY": "",
      "NIMBUS_MCP_KEY_FILE": "",
      "MCP_LOCAL_KEY": "",
      "MCP_LOCAL_USER_ID": "",
      "HOST": ""
    }
  }
}
```

**Requires environment variables:** `NIMBUS_API_URL`, `NIMBUS_TOKEN`, `NIMBUS_TOKEN_FILE`, `NIMBUS_MCP_KEY`, `NIMBUS_MCP_KEY_FILE`, `MCP_LOCAL_KEY`, `MCP_LOCAL_USER_ID`, `HOST` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What Nimbus BCI does

The Nimbus BCI MCP server exposes Nimbus BCI workflows to MCP-compatible agents. It covers the path from data and pipeline construction through execution, model training, evaluation, visualization, and experiment analysis. Agents can upload EEG data, work with public MOABB data packs, save pipelines into Studio projects, and run campaigns across multiple configurations.

The server also supports live EEG sessions. Agents can start and monitor streams, read telemetry, inspect signal quality, and obtain modelless focus or relaxation indicators. Guided subject calibration is available through the calibration tools. The system can drive supported hardware through a local backend, including BrainFlow boards, BrainBit over BLE, LSL, and PiEEG, as well as synthetic boards. The material does not describe stimulation output, spike or intracortical ephys data, source localization, or full MNE feature parity.

## How it works

Nimbus BCI connects an MCP client to either a Nimbus backend running on the same machine or the hosted Nimbus deployment. The package provides a stdio server, while the hosted setup also uses an HTTP gateway. A client sends tool calls for account information, data, pipelines, executions, plots, streaming, calibration, BIDS export, and grouped statistics.

Authentication can use a device-code login, credentials discovered from the Nimbus Studio desktop app, or environment-provided credentials. With device login, `nimbus-mcp login` opens an approval flow and stores the resulting credential locally. A hosted token runs requests under the associated Nimbus account, so account quotas and plan limits apply. Hosted token-authenticated runs are CPU-only in version 0.4.

Without a credential, the server still starts in setup mode. Tool calls return onboarding information instead of failing at process startup. The `account.whoami` tool reports identity, plan details, free-run quota, token status, and credential source.

## Setup and configuration

Install the package with `pip install nimbus-mcp` or run it with `uvx nimbus-mcp`. The repository also supports an editable install from `nimbus-studio/mcp`.

For hosted use, run `nimbus-mcp login`; `nimbus-mcp status` checks the credential, plan, quota, and backend connection, while `nimbus-mcp logout` removes the stored credential. Advanced or CI deployments can use `NIMBUS_TOKEN`, `NIMBUS_TOKEN_FILE`, `NIMBUS_MCP_KEY`, or `NIMBUS_MCP_KEY_FILE`.

Local use requires a Nimbus backend, either through the desktop application or the development server. The backend must be configured with `MCP_LOCAL_KEY`; `MCP_LOCAL_USER_ID` identifies the owner of projects, pipelines, and executions created by the agent. A headless backend can operate supported hardware without the desktop application when the streaming dependencies are installed.

## Tools and capabilities

The Nimbus BCI MCP server supports capabilities including:

- Building, validating, and running BCI pipelines.
- Training on uploaded EEG data and MOABB packs.
- Supplying custom `python_model` code in local or cloud sandbox execution.
- Running multi-configuration experiments with leaderboard and server-side LOSO scoring.
- Returning agent-visible PNG figures through `plots.*` tools.
- Starting, monitoring, and inspecting live streams through `stream.start`, `stream.status`, and `stream.telemetry`.
- Calibrating subjects, exporting BIDS data with SHA-256 manifests, and calculating grouped statistics.

Streaming sessions have an inactivity watchdog that stops sessions after 15 minutes without a status or telemetry poll by default. Passing `idle_timeout_sec=0` disables that timeout for a session.

## Limitations and notes

The server does not send stimulation events, handle spike or intracortical ephys data, provide source localization, or claim complete MNE coverage. Local hardware access depends on the backend and its device dependencies. Hosted execution consumes the connected account's quota, and tokens expire after a maximum of 90 days, with 30 days as the default lifetime. A plan change may require a new login or token before the updated plan is reflected.

The Nimbus BCI MCP server is best suited to agent-driven Biology, Medicine and Bioinformatics workflows centered on Nimbus pipelines, EEG experiments, and live BCI sessions. It is less appropriate when the primary requirement is stimulation control, intracortical recording analysis, or a general-purpose offline MNE interface.

_Full upstream README: https://allmcps.com/mcp/nimbus-bci/readme_