Ros MCP Server

What ros-mcp-server MCP server does

The ros-mcp-server MCP server exposes ROS robot systems to large language model clients through the Model Context Protocol. It is intended for both physical robots and simulated systems, and supports ROS 1 as well as ROS 2 distributions such as Jazzy and Humble.

The connection is bidirectional. An agent can inspect robot data and state, then send commands through the ROS interfaces already present in the system. The project describes use cases including industrial end-effector diagnosis, natural-language control of a mobile manipulator, and control of a simulated Unitree Go2 in NVIDIA Isaac Sim.

No changes to existing robot application code are required according to the project documentation. The ROS setup uses a rosbridge node, allowing the MCP layer to communicate with the robot through the existing ROS environment.

How it works

An MCP-compatible client connects to the server and provides an AI model with access to ROS interfaces. The server can discover topics, services, actions, and their message or request types. This discovery includes custom types, which lets an agent determine the syntax needed to interact with a particular robot rather than relying only on predefined configuration.

The agent can then use the discovered interfaces to read sensor information, monitor state, publish or subscribe to topics, call services and actions, and set parameters. The exact behavior available to the model depends on the interfaces exposed by the connected ROS system.

The ros-mcp-server MCP server is suitable for systems where an agent needs both current robot context and the ability to issue ROS operations. It does not replace the robot's navigation, manipulation, simulation, or other ROS components; those capabilities remain part of the connected system.

Setup and configuration

Installation is documented in the repository's installation guide. The project requires Python 3.10 or newer and pip 23.0 or newer. A development-container setup is also indicated by the repository badges.

A ROS environment with rosbridge must be available for the server to connect to an existing robot setup.

The server can be used with Claude Code, Codex CLI, Gemini CLI, Claude Desktop, ChatGPT, Cursor, and other MCP-compatible clients.

Tools and capabilities

The ros-mcp-server MCP server supports the following ROS interaction categories:

• Discover available topics, services, actions, and their types.
• Publish messages to ROS topics.
• Subscribe to topics and inspect incoming data.
• Call ROS services and actions.
• Read sensor data and monitor robot state.
• Set ROS parameters.
• Work with custom ROS interface types discovered from the connected system.

These capabilities support tasks such as testing an industrial robot gripper, investigating an anomaly using robot data and manuals, navigating a robot, and commanding a manipulator when the underlying ROS stack provides those functions.

Limitations and notes

The repository lists permissions as a possible future feature, so deployments should account for the risks of allowing an AI model to invoke control interfaces.

Robot behavior depends on the connected ROS graph and its existing nodes. For example, navigation and manipulation demonstrations rely on components such as Nav2, SLAM, and MoveIt in the robot system; the MCP server itself does not claim to provide those subsystems.

The project is licensed under Apache License 2.0.