Provides text-to-speech capabilities through Deepgram's API, enabling the Reachy Mini robot to speak using the speak tool.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Reachy Mini MCP show me a curious expression"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
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/
Token-efficient tool reference for programmatic use:
Tool | Args | Purpose |
|
| Voice + gesture, optionally listen after |
|
| STT via Deepgram Nova-2 |
| - | Camera capture (base64 JPEG) |
|
| Express emotion or play recorded move |
|
| Head positioning (degrees) |
|
| neutral / sleep / wake |
|
| Find available recorded moves |
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.
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.
Related MCP server: Reachy Mini MCP Server
# 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.pyAI (Claude/GPT/Grok) → MCP Server → SDK → Daemon → Robot/Simulator7 tools following Miller's Law—fits in working memory.
Provider | Status | Use Case |
✅ Supported | xAI's expressive voice (Eve, Ara, Leo, Rex, Sal) | |
✅ Supported | TTS (Aura 2) + STT (Nova 2) |
Grok Voice is used automatically when XAI_API_KEY is set. Falls back to Deepgram otherwise.
~/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.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"
}
}
}
}Variable | Required | Default | Purpose |
| No | - | Grok Voice TTS (preferred) |
| No |
| Grok voice: ara, eve, leo, rex, sal |
| Yes* | - | STT (always required for listen) + TTS fallback |
| No |
| Daemon API endpoint |
*Required for listen(). Also required for speak() if XAI_API_KEY not set
Python 3.10+
reachy-mini SDK (installed via poetry)
MuJoCo (for simulation)
Deepgram API key (for speak/listen)
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
MIT License - see LICENSE
Reachy Mini SDK by Pollen Robotics (Apache 2.0)
Grok Voice integration pattern from dillera's reachy_mini_conversation_app
Reachy Mini SDK (Apache 2.0)
This server cannot be deployed
Robot embodied-AI cloud for GPU training, inference, benchmarks, and robot data collection.
Connect Claude, Cursor, Codex, and other AI tools to your robotics mission data.
Operate Linux, macOS and Windows from your LLM. Every action runs through an auditable allowlist.
A comprehensive Model Context Protocol (MCP) server that enables AI assistants to control Unreal E…
