Unitree Go2 MCP Server

What lpigeon/unitree-go2-mcp-server MCP server does

lpigeon/unitree-go2-mcp-server MCP server provides an MCP interface for controlling a Unitree Go2 robot. An AI system receives a natural-language request, interprets the intended action, and passes the resulting command to the server. The server then translates that request into ROS 2 instructions used by the Go2 environment.

The README demonstrates a movement request that asks the robot to travel forward at a specified velocity for a specified duration. It also describes contextual interaction, such as interpreting a comment that the robot appears tired, and a broader demonstration involving obstacle avoidance, changing direction, and user interaction. The available MCP functions are documented separately in the repository's MCPFUNCTIONS.md file.

How it works

The server depends on the unitree_ros2 environment supplied by Unitree Robotics. ROS 2 provides the communication layer between the MCP process and the robot. Before using the server, the setup of unitree_ros2 must be completed through its “Connect and test” step.

An MCP-compatible AI application launches the server with uv and server.py. The application sends the user request through MCP, while the server uses the configured ROS 2 setup script to access the robot. The README instructs operators to verify connectivity with ros2 topic list and look for the /wirelesscontroller topic.

Setup and configuration

Run the server on the computer connected to the Go2. The documented setup requires Ubuntu 20.04 or 22.04, a Unitree Go2 robot, and ROS 2 Humble or Foxy; Humble is recommended.

Clone the repository, install uv, and optionally create a virtual environment. Add an MCP server entry that invokes uv, points --directory to the repository, and runs server.py. The configured directory must use an absolute path.

Before running an AI client, edit server.py and set UNITREE_ROS2_SETUP_SH_PATH to the location of the Unitree ROS 2 setup script. If using rosbridge, also configure the local IP, rosbridge IP, and rosbridge port. The documented default port is 9090.

Tools and capabilities

The lpigeon/unitree-go2-mcp-server MCP server exposes MCP functions for requesting actions from the Go2 through the configured ROS 2 environment.

The documented usage covers natural-language robot control, movement with velocity and duration parameters, contextual requests, and a demonstration combining obstacle avoidance, direction changes, and user interaction. These capabilities depend on the robot being reachable through the configured network and ROS 2 setup.

Limitations and notes

This server is specifically tied to the Unitree Go2 and its ROS 2 environment. It is not presented as a general robot-control interface. The configuration must be performed on the PC connected to the Go2, and the robot must be connected to the network before commands are sent.

If /wirelesscontroller does not appear in the output of ros2 topic list, the README directs the operator to check the robot's network connection. The server also requires local path configuration in server.py; the supplied instructions do not describe environment variables or automatic discovery for these settings.

Natural-language behavior depends on the LLM or AI application importing the server. The examples show contextual interpretation, but the repository excerpt does not specify model requirements, safety controls, or the complete MCP function catalog.