BrainFlow Biosensing Boards
io.github.K-Dense-AI/labmcp-brainflow · 19 stars · Python · Apache-2.0
MCP server MCP server for EEG/EMG/ECG/PPG boards via BrainFlow (OpenBCI, Muse, Neurosity, Unicorn, BrainBit).
Install
The repo has no one-line install. Follow its README.
Published as
- PyPI
labmcp-brainflowstdio
From the server's entry in the official MCP Registry.
Configuration
| Name | Set as | What it is |
|---|---|---|
LABMCP_ADDRESSOptional | Env var | Instrument address, e.g. serial:///dev/ttyUSB0 or tcp://192.168.1.50:5025 |
LABMCP_SIMULATEOptional | Env var | Set to 1 to use the built-in simulator (no hardware) |
LABMCP_READ_ONLYOptional | Env var | Set to 1 to disable all state-changing tools |
LABMCP_LIMITSOptional | Env var | Safety limit overrides, e.g. max_temperature_c=80 |
Files
BrainFlow Biosensing Boards — MCP Server
Let an AI agent stream, record and analyse EEG, EMG, ECG, EOG, PPG and EDA from the biosensing boards supported by the open-source BrainFlow SDK: OpenBCI, Muse, Neurosity, g.tec Unicorn, BrainBit and many more. It covers recording with event markers, EEG band powers and per-electrode signal-quality checks.
Research use only. This is not a medical device. Do not use it to diagnose, monitor or treat anyone. Follow the board manufacturer's safety instructions (battery operation, electrical isolation, skin preparation). This server does not change any of them.
| Package | labmcp-brainflow |
| Instruments | OpenBCI Cyton, Cyton+Daisy, Ganglion, Galea (USB dongle, BLE, WiFi Shield) · Muse 2 / S / S Athena / 2016 · Neurosity Crown / Notion · g.tec Unicorn · BrainBit · Callibri · Enophone · Mentalab Explore · EmotiBit · FreeEEG32/128 · NeuroPawn Knight · Shimmer3 · any other board in BrainFlow's BoardIds |
| Interfaces | USB serial dongle, Bluetooth LE (native or BLED112 dongle), WiFi, vendor SDKs (all through BrainFlow) |
| Protocol | BrainFlow SDK ≥ 5.21: BoardShim, BrainFlowInputParams, DataFilter (User API, Supported Boards, Data Format) |
| Status | 🧪 simulated: tested against BrainFlow's own synthetic board and a built-in fake, not yet verified on hardware. Report a hardware test |
Try it without hardware
uvx labmcp-brainflow --simulate --check
--simulate runs BrainFlow's own SYNTHETIC_BOARD, so the real BrainFlow code path runs: 16 EXG channels at 250 Hz with sine waves at 5, 10, 15 … 80 Hz plus accelerometer, gyro, PPG, EDA and temperature rows. On platforms where BrainFlow's native library can't load, or with --option simulator=fake, a pure-numpy stand-in is used instead. It simulates an 8-channel, 250 Hz EEG headset with 1/f background, posterior alpha at 10 Hz and one badly attached electrode (Fp2, heavy 50 Hz pickup), so the signal-quality check has something to find.
Connect your board
- Pick the board with
--option board=<alias>(or any BrainFlowBoardIdsname such asANT_NEURO_EE_411_BOARD, or a numeric id). Thelist_supported_boardstool prints this table too. - Give the connection detail with
--address. Depending on the board, BrainFlow expects a differentBrainFlowInputParamsfield, and the server fills in the right one:
Board (--option board=) |
--address is |
Required? | Notes |
|---|---|---|---|
cyton, cyton_daisy |
serial port of the USB dongle | yes | macOS: use /dev/cu.usbserial-*, not /dev/tty.* |
ganglion (native BLE) |
Bluetooth MAC | no (auto-discovery) | FW 2 boards: --option other_info=fw:2 |
ganglion_dongle |
serial port of the BLED112 dongle | yes | optional --option mac_address= |
cyton_wifi, cyton_daisy_wifi, ganglion_wifi |
WiFi Shield IP (192.168.4.1 in direct mode) | no (SSDP) | --option ip_port= (default 6789) |
muse_2, muse_s, muse_s_athena, muse_2016 |
Bluetooth MAC | no | --option serial_number=Muse-XXXX; PPG: --option other_info=p50 (Muse 2) / p61 (Muse S) |
muse_2_bled, muse_s_bled |
serial port of the BLED112 dongle | yes | |
crown, notion_1, notion_2 |
device serial number | no | same network, broadcast must be allowed |
unicorn, brainbit |
device serial number | no | Unicorn: pair with the supplied dongle |
enophone, explore_4, explore_8 |
Bluetooth MAC | Linux only | |
emotibit |
network broadcast address | no | |
freeeeg32, freeeeg128, neuropawn_knight, shimmer3, ironbci_32 |
serial port | yes | |
playback |
path of a file saved in BrainFlow format | yes | --option master_board=<board it was recorded with> |
synthetic |
nothing | BrainFlow's synthetic board |
For a board that isn't listed, the server infers the field from the address format: COM3 or /dev/… becomes a serial port, AA:BB:… a MAC address, a.b.c.d an IP address, and anything else a serial number. You can also set any field directly with --option serial_port=… / mac_address=… / ip_address=… / serial_number=… / other_info=… / timeout=….
- Test the connection:
uvx labmcp-brainflow --option board=cyton --address /dev/cu.usbserial-DM00Q0QN --check # macOS uvx labmcp-brainflow --option board=cyton_daisy --address COM3 --check # Windows uvx labmcp-brainflow --option board=muse_2 --option serial_number=Muse-1A2B --check uvx labmcp-brainflow --option board=unicorn --check
Close other programs that hold the device (OpenBCI GUI, Muse app) first. On Linux, add yourself to thedialoutgroup for serial ports.
Add to your MCP client
Claude Code
claude mcp add eeg -- uvx labmcp-brainflow --option board=cyton --address /dev/ttyUSB0
Claude Desktop / Cursor / Windsurf (claude_desktop_config.json, .cursor/mcp.json, …)
{
"mcpServers": {
"eeg": {
"command": "uvx",
"args": ["labmcp-brainflow", "--option", "board=cyton", "--address", "/dev/ttyUSB0"]
}
}
}
With --read-only, the agent can still record and analyse data (a temporary stream is started for each recording) but cannot leave the stream running, insert markers or send configure_board commands. stop_streaming stays available.
Tools
| Tool | Kind | Description |
|---|---|---|
configure_board |
⚠️ hazard | Send a raw board-specific command to the firmware through BrainFlow's config_board (e.g. OpenBCI channel settings 'x1060110X', test signals, or Muse presets 'p50'/'p61' to enable PPG). |
get_band_powers |
👁 read | EEG band powers (delta 1-4, theta 4-8, alpha 8-13, beta 13-30, gamma 30-50 Hz) over the most recent window_s seconds: BrainFlow's channel-averaged relative powers plus per-channel absolute (uV^2) and relative powers and the peak frequency. Records a fresh window if not streaming. |
get_board_info |
👁 read | Describe the connected board: channel names by type (EEG/EMG/ECG/EOG share the EXG rows on most boards), sampling rate, available presets (data buffers) and streaming state. |
get_command_log |
👁 read | Return the most recent raw commands sent to / replies received from the instrument (newest last). Useful for debugging and for recording what was done. |
get_connection_info |
👁 read | Report which instrument is connected (identity, address, simulated or real), whether the server is read-only, and the active safety limits. Call this first. |
get_signal_quality |
👁 read | Check every EXG channel for common electrode problems: flat line (disconnected), railed (amplifier saturated, OpenBCI Cyton boards), strong 50/60 Hz mains noise (poor contact or missing reference), and implausibly high amplitude (movement, muscle, loose electrode). |
insert_marker |
🎛 control | Write an event marker into the data stream at the current sample (for event-related experiments: stimulus onsets, condition changes). Requires start_streaming; markers appear in record results and saved files. |
list_supported_boards |
👁 read | List common BrainFlow boards: the --option board= alias, BrainFlow board id, and which connection detail --address must hold (serial port, Bluetooth MAC, IP address or serial number). Any other BrainFlow BoardIds name or numeric id is accepted too. Does not need a board. |
reconnect |
🛑 safety | Close and re-open the connection to the instrument (e.g. after it was power cycled or a cable was re-plugged). |
record |
👁 read | Record duration_s seconds and return per-channel statistics, event markers and downsampled traces. Uses the live stream if one is running, otherwise starts a temporary one. The full data (every row, full sampling rate) can be written to save_path. stop_streaming ends a recording early. |
start_streaming |
🎛 control | Start continuous acquisition into BrainFlow's ring buffer (the board's radio/LEDs switch on; nothing is applied to the participant). Needed for insert_marker; record then reads from the live stream. Call stop_streaming when finished. |
stop_streaming |
🛑 safety | Stop acquisition (saves battery). A record in progress ends at once with the data acquired so far. Data already in the buffer is kept until the next stream starts or the session is released (reconnect). |
Safety limits
| Limit | Default | Meaning |
|---|---|---|
max_record_duration_s |
60 s | Longest recording or analysis window (record, get_band_powers, get_signal_quality) an agent may request |
Override at launch: --limit max_record_duration_s=600. BrainFlow boards measure; the only tool that can switch on an output (for example an impedance-test current through the electrodes, or the bias drive on a channel) is configure_board, a raw firmware passthrough that is marked as a hazard so MCP clients ask before running it.
Example prompts
- "Check the signal quality on all 8 Cyton channels and tell me which electrodes need more gel."
- "Record 30 s eyes-closed and 30 s eyes-open, save both as CSV in ~/data/sub-01/, and compare occipital alpha power."
- "Start streaming. I'll run an oddball task: insert marker 1 for standards and 2 for targets when I tell you. Then save the session in BrainFlow format."
- "What's the peak alpha frequency on O1 and O2 over the last 10 seconds?"
- "Is there 50 or 60 Hz mains noise in this setup, and on which channels?"
- "List the boards you support and tell me how to connect my Muse S."
Notes
- How the band powers are computed.
average_relativecomes from BrainFlow'sDataFilter.get_avg_band_powers(apply_filter=True): detrend, 48–52 and 58–62 Hz band-stop, 2–45 Hz band-pass, Welch PSD, then bands of 1–4, 4–8, 8–13, 13–30 and 30–50 Hz averaged over channels and normalised to sum to 1. Per-channel values use detrend, a 50/60 Hz notch and a 1–45 Hz 4th-order zero-phase Butterworth (BrainFlowDataFilter), followed by a Welch PSD (Hann window, 50 % overlap, ≈0.5 Hz bins). Absolute band power is the integrated PSD in µV², and gamma is effectively 30–45 Hz. - Signal-quality verdicts (EEG-oriented thresholds):
flatmeans the raw SD is below 0.5 µV (disconnected);railedmeans at least 90 % of the ADS1299 input range is used (OpenBCI Cyton family only, using BrainFlow's formula with gain 24, set by--option exg_gain=);line_noisemeans 50 or 60 Hz power exceeds the whole 1–45 Hz power;high_amplitudemeans the band-passed RMS is above 100 µV. EMG/ECG recordings will naturally be "high amplitude". - Units. BrainFlow returns EXG in µV "wherever possible". The OYMotion gForce armbands return ADC counts. Accelerometer, PPG and similar channels use board-specific units. Timestamps are Unix seconds.
- Presets. Some boards stream several buffers with different sampling rates, for example Muse EEG (default), IMU (auxiliary) and PPG (ancillary). Pass
preset=toget_board_infoandrecord. - Markers go into the default preset's marker row and need an active stream (
start_streaming). You can insert them whilerecordis running. Value 0 is reserved. - Saved files.
save_pathmust be a new file (.csv;.csv,.tsvor.txtforsave_format="brainflow"):~is expanded, missing folders are created, an existing file is never overwritten, and the path is checked before recording starts.save_format="csv"writes labelled columns for every row.save_format="brainflow"writesDataFilter.write_fileoutput, which you can replay with--option board=playback --address file.csv --option master_board=<board>. configure_boardpasses a string straight to the board firmware through BrainFlow'sconfig_board, for example OpenBCI channel settings or Muse presets. Nothing is checked: some commands switch on outputs such as the lead-off (impedance-test) current. Settings persist until changed or the board is power-cycled; check the board's SDK documentation for valid commands.- Streams.
record,get_band_powersandget_signal_qualitystart a temporary stream when none is running and stop it afterwards, unlessstart_streamingwas called meanwhile (the stream then stays on) or another recording still uses it.stop_streamingends a runningrecordat once; it returns only the data acquired so far, withstopped_early: true. reconnect(and stopping the server) stops the stream and callsrelease_session, which frees the serial port or BLE connection.
Hardware verification
| Model | Firmware | Interface | Verified by | Date |
|---|---|---|---|---|
| none yet: be the first |
Facts
- Kind
- MCP server
- Group
- Uncategorized
- Stars
- 19
- License
- Apache-2.0
- Language
- Python
- Last push
- 2026-09-26
- Forks
- 1
- MCP Registry
- io.github.K-Dense-AI/labmcp-brainflow
- Homepage
- www.k-dense.ai
- Topics
- ai-agents, instrument-control, lab-automation, laboratory, mcp, mcp-server, model-context-protocol, python, scientific-instruments, self-driving-labs
- 1REAio.github.morluto/reaReverse engineer anything from your terminal or agent with one CLI and MCP server.125.3k
- 2com.browser-use/browser-usecom.browser-use/browser-useControl a real Chrome browser to complete any task: fill forms, extract data, book flights.117.7k
- 3World Monitorapp.worldmonitor/mcpLive markets, conflicts, country risk, chokepoints, energy, and China decision signals. 93 tools.88.3k
- 4Scrapling MCP Serverd4vinci/scrapling🕷️ An adaptive Web Scraping framework that handles everything from a single request to a full-scale crawl! Don't be shy, join here: https://discord.gg/EMgGbDceNQ and follow here for daily tips and tricks: https://x.com/Scraplingdev86.9k
- 5Everythingmodelcontextprotocol/serversThis MCP server attempts to exercise all the features of the MCP protocol. It is not intended to be a useful server, but rather a test server for builders of MCP clients. It implements prompts, tools, resources, sampling, and more to showcase MCP capabilities.85.8k
- 6Fetchmodelcontextprotocol/serversA Model Context Protocol server that provides web content fetching capabilities. This server enables LLMs to retrieve and process content from web pages, converting HTML to markdown for easier consumption.85.8k