Reachy Mini MCP

01 · What it does

• speak: voice plus gesture, with move markers embedded in the text
• listen: speech to text through Deepgram
• snap: capture a camera frame
• show: one of 12 built-in emotions or one of 81 recorded moves from Pollen
• look and rest: head roll, pitch, yaw and height, and neutral, sleep or wake poses

02 · Install

$ git clone https://github.com/jackccrawford/reachy-mini-mcp.git
$ cd reachy-mini-mcp
$ poetry install
$ export DEEPGRAM_API_KEY=your_key_here
$ poetry run python src/server.py

1. The README starts at cd reachy-mini-mcp, so the clone line uses the repo URL
2. Simulator: mjpython -m reachy_mini.daemon.app.main --sim --scene minimal. On real hardware the same server connects to the daemon
3. Set XAI_API_KEY to use Grok Voice for speech; otherwise Deepgram is used. Deepgram is always needed for listen
4. Daemon at http://localhost:8321/api by default (REACHY_DAEMON_URL); Zenoh uses port 7447

03 · Client config (from README)

{
  "mcpServers": {
    "reachy-mini": {
      "command": "poetry",
      "args": ["-C", "/path/to/reachy-mini-mcp", "run", "python", "src/server.py"],
      "env": {
        "DEEPGRAM_API_KEY": "your_key_here"
      }
    }
  }
}

Install steps taken from the project README (its recommended path first). Placeholders are the maintainer's. Not bench-tested by HardwareMCP yet.