Reachy Mini MCP

Give your AI a body.

This MCP server lets AI systems control Pollen Robotics' Reachy Mini robot—speak, listen, see, and express emotions through physical movement. Works with Claude, GPT, Grok, or any MCP-compatible AI.

7 tools. 30 minutes to first demo. Zero robotics expertise required.

https://reachy-mini-mcp-969sxyq.gamma.site/

For AI Systems

Token-efficient tool reference for programmatic use:

Tool	Args	Purpose
`speak`	`text, listen_after=0`	Voice + gesture, optionally listen after
`listen`	`duration=3`	STT via Deepgram Nova-2
`snap`	-	Camera capture (base64 JPEG)
`show`	`emotion, move=""`	Express emotion or play recorded move
`look`	`roll, pitch, yaw, z, duration`	Head positioning (degrees)
`rest`	`mode="neutral"`	neutral / sleep / wake
`discover`	`library="emotions"`	Find available recorded moves

speak()

Supports embedded move markers for choreographed speech:

speak("[move:curious1] What's this? [move:surprised1] Oh wow!")

Moves fire right before their speech chunk. Use listen_after=5 to hear response.

show()

Built-in emotions (fast, local):
neutral, curious, uncertain, recognition, joy, thinking, listening, agreeing, disagreeing, sleepy, surprised, focused

Recorded moves (81 from Pollen):

show(move="loving1")
show(move="fear1")
show(move="serenity1")

Use discover() to see all available moves.

Quick Start

# Install
cd reachy-mini-mcp
poetry install

# Set API key (required for speak/listen)
export DEEPGRAM_API_KEY=your_key_here

# Start simulator daemon
mjpython -m reachy_mini.daemon.app.main --sim --scene minimal

# Run MCP server
poetry run python src/server.py

Architecture

AI (Claude/GPT/Grok) → MCP Server → SDK → Daemon → Robot/Simulator

7 tools following Miller's Law—fits in working memory.

Voice Providers

Provider	Status	Use Case
Grok Voice	✅ Supported	xAI's expressive voice (Eve, Ara, Leo, Rex, Sal)
Deepgram	✅ Supported	TTS (Aura 2) + STT (Nova 2)

Grok Voice is used automatically when XAI_API_KEY is set. Falls back to Deepgram otherwise.

MCP Config

Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json:

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

Claude Code

~/.claude.json:

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

Environment Variables

Variable	Required	Default	Purpose
`XAI_API_KEY`	No	-	Grok Voice TTS (preferred)
`GROK_VOICE`	No	`eve`	Grok voice: ara, eve, leo, rex, sal
`DEEPGRAM_API_KEY`	Yes*	-	STT (always required for listen) + TTS fallback
`REACHY_DAEMON_URL`	No	`http://localhost:8321/api`	Daemon API endpoint

*Required for listen(). Also required for speak() if XAI_API_KEY not set

Requirements

Python 3.10+
reachy-mini SDK (installed via poetry)
MuJoCo (for simulation)
Deepgram API key (for speak/listen)

Hardware Notes

Simulator: mjpython required on macOS for MuJoCo visualization
Real hardware: Same MCP server, daemon auto-connects
Port conflicts: Zenoh uses 7447, daemon uses 8321 by default

License

MIT License - see LICENSE

Acknowledgments

Reachy Mini SDK by Pollen Robotics (Apache 2.0)
Grok Voice integration pattern from dillera's reachy_mini_conversation_app

reachy-mini-mcp