EEG Data MCP Server

Overview

This is the connective tissue between a brain and a language model. The MindWave capture pipeline produces clean EEG; this MCP server makes that stream queryable -; it connects to the headset over Bluetooth, runs FFT for the frequency bands, and presents the whole thing as a set of MCP tools so an AI can ask "what's my brain doing right now?" and get a structured answer.

It's one of several MCP servers in the collection, and the most physically grounded of them -; the only one whose input is electrodes on a forehead.

Background

Once the raw MindWave capture worked, the obvious move was to stop dumping CSVs and start feeding the data to a model in real time. MCP was the right shape for that: a clean tool interface, discoverable from Claude's tools menu, that hides all the Bluetooth-and-FFT plumbing behind a handful of verbs.

How It Works

The server wraps the existing MindWave codebase -; reusing the Bluetooth connection handler and ThinkGear parser -; and adds an async streaming layer on top:

• Tools -; get_eeg_status (connection + signal quality), get_current_data (latest readings), start_recording / stop_recording (CSV capture), and get_stream (live data).
• Frequency analysis -; a circular buffer holds the last second of raw data so SciPy can compute band power continuously.
• Robustness -; automatic reconnection with exponential backoff, overflow-protected buffering, and active signal-quality monitoring that flags when the headset needs adjusting.
• Derived metrics -; flow-state score, cognitive load, focus stability, alertness, and a stress indicator, computed from band ratios on top of the raw stream.

The design trims raw-sample transmission (sending roughly every tenth sample) to avoid flooding the protocol while keeping the frequency analysis accurate.

Where It Landed

Archived as a working server -; there are real run logs from it bridging the headset to Claude. It does what it set out to do; it just isn't something I left running day to day.

• Live EEG streaming exposed through a clean MCP tool surface.
• Continuous FFT band analysis plus derived cognitive-state metrics.
• Reuses the MindWave capture stack rather than reimplementing connection handling.