ros-mcp

Connect AI agents to ROS robots. Exposes topics, services, nodes, parameters, and actions as MCP tools — usable from Claude Desktop, Claude Code, Cursor, or any MCP-compatible client.

Mode	Best for
Browser dashboard	Quickest start, no install
Python server — stdio	Claude Code (local)
Python server — HTTP	Claude Desktop, Cursor, remote agents
Remote CI server	Shared / cloud-hosted setup

How it works

The browser dashboard skips the Python server entirely — roslibjs connects directly to rosbridge from the browser, with an embedded AI chat panel.

Quick start

git clone https://github.com/jonasneves/ros-mcp
cd ros-mcp

Browser dashboard — open ros-mcp.github.io, enter your rosbridge WebSocket URL, and connect. No install required.

Claude Code (stdio)

make configure

Claude Desktop / Cursor (HTTP on :9000)

ROSBRIDGE_IP=<robot-ip> make server-http
make configure-desktop

Remote CI server

make configure-remote

Docker simulators

3× Turtlesim — simulators and rosbridge start automatically, MCP server at http://localhost:9000/mcp:

make turtlesim

Isaac Sim — requires Linux with NVIDIA GPU, Isaac Sim installed, and the ROS 2 Bridge extension enabled:

make isaac-sim                           # rosbridge (FastDDS, host network)
ROSBRIDGE_IP=127.0.0.1 make server-http  # MCP server, separate terminal

On macOS or a remote machine, point directly at a rosbridge already running alongside Isaac Sim: ROSBRIDGE_IP=<host-ip> make server-http.

Tools

30+ tools covering topics, services, nodes, actions, parameters, robot description, joint states, images, and motion control. See docs/tools.md for the full reference.

Configuration

Variable	Default	Description
ROSBRIDGE_IP	127.0.0.1	rosbridge host
ROSBRIDGE_PORT	9090	rosbridge port
ROS_DEFAULT_TIMEOUT	5.0	Timeout in seconds

License

Apache 2.0. See LICENSE.