ROS MCP

Overview

The ROS MCP Server is designed to support robots in performing complex tasks and adapting effectively to various environments by providing a set of functions that transform natural language commands, entered by a user through an LLM, into ROS commands for robot control. Furthermore, by utilizing rosbridge, it is configured to operate with both ROS and ROS2 systems, and its WebSocket-based communication enables broad applicability across diverse platforms.

Research based on this project can be found in the video linked below.

• An Efficient Robot Control Framework Using the Model Context Protocol

Supported Types

• geometry_msgs/Twist
• sensor_msgs/Image
• sensor_msgs/JointState

Features

• WebSocket-based universal compatibility: Communicates with both ROS and ROS2 systems using rosbridge, enabling seamless integration regardless of ROS version.
• Cross-platform support: Works on Linux, Windows, and MacOS, making it suitable for diverse development and deployment environments.
• Easy integration with LLMs and AI systems: Natural language commands can be directly translated into robot actions via MCP functions.
• Extensible function set: Easily add new robot control or sensor functions by extending the MCP tool interface.
• No ROS node modification required: Interacts with existing ROS/ROS2 topics and services without changing your robot's core code.
• Native ROS/ROS2 command compatibility: Optionally supports using local ROS/ROS2 libraries, so you can run native ROS commands and tools alongside WebSocket-based control.

Contributing

Contributions are welcome!
Whether you're fixing a typo, adding a new function, or suggesting improvements, your help is appreciated.
Please follow the contributing guidelines for more details on how to contribute to this project.

Installation

Installing via Smithery

To install ros-mcp-server for Claude Desktop automatically via Smithery:

npx -y @smithery/cli install @lpigeon/ros-mcp-server --client claude

Installing Locally

uv Installation

• To install uv, you can use the following command:

curl -LsSf https://astral.sh/uv/install.sh | sh

or

pip install uv

• Create virtual environment and activate it (Optional)

uv venv
source .venv/bin/activate

MCP Server Configuration

Set MCP setting to mcp.json.

{
  "mcpServers": {
    "ros-mcp-server": {
      "command": "uv",
      "args": [
        "--directory",
        "/ABSOLUTE/PATH/TO/PARENT/FOLDER/ros-mcp-server",,
        "run",
        "server.py"
      ]
    }
  }
}

If you use Claude Desktop, you can find mcp.json using the following command:

• MacOS

code ~/Library/Application\ Support/Claude/claude_desktop_config.json

• Linux(Ubuntu)

You can install Claude Desktop to use claude-desktop-debian.

code ~/.config/Claude/claude_desktop_config.json

• Windows

code $env:AppData\Claude\claude_desktop_config.json

MCP Functions

You can find the list of functions in the MCPFUNCTIONS.md.

How To Use

1. Set IP and Port to connect rosbridge.

• Open server.py and change your LOCAL_IP, ROSBRIDGE_IP and ROSBRIDGE_PORT. (ROSBRIDGE_PORT's default value is 9090)

2. Run rosbridge server.

ROS 1

roslaunch rosbridge_server rosbridge_websocket.launch

ROS 2

ros2 launch rosbridge_server rosbridge_websocket_launch.xml

3. Run any AI system that has imported ros-mcp-server.

4. Type "Make the robot move forward.".

5. Check rosbridge_server and ros topic.

• rosbridge_server

• ros topic

Simulation Test

MCP-based control using the MOCA mobile manipulator within the NVIDIA Isaac Sim simulation environment.