robotmcp_client

The ROS MCP Client is a reference MCP client that connects to a ROS MCP server to enable natural-language interactions with ROS/ROS2 robots via the MCP protocol. It acts as the LLM-side client, streaming queries and responses between your local language model and the robot ecosystem through ros-mcp-server. Key capabilities include bidirectional streaming of commands and sensor data, ROS-aware prompts and handlers, and integration with local LLMs such as Gemini, Ollama, or Nvidia NeMo. To use it, first ensure a ros-mcp-server instance is running and accessible, and that rosbridge is available on your robot so the MCP bridge can communicate with ROS topics and services. You can then run the client locally and connect to the server over MCP (stdio or HTTP) to start issuing natural-language or structured requests to the robot, receiving streamed feedback and robot state data as responses. The client includes a Gemini Live Client integration example under clients/gemini_live for Gemini-based reasoning, and it supports offline operation in LAN-only setups when needed.

Prerequisites:

• Python 3.10+ installed on your machine
• ROS or ROS2 environment with rosbridge available
• Internet access for initial install

Installation steps:

1. Clone the repository

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

2. Install dependencies via uv (uvx) or pip in editable mode

uv sync  # or: pip install -e .

3. Set up the Gemini Live client if you intend to use Gemini as the reasoning engine

• Follow the setup in clients/gemini_live/README.md to configure Gemini integration

4. Start rosbridge on the target robot (example for ROS2)

ros2 launch rosbridge_server rosbridge_websocket_launch.xml

5. Run the MCP client and connect to the ROS MCP server

uv run ros-mcp-client

Note: Adjust the command to point to the correct entry script if your environment differs.

Environment and configuration tips:

• Ensure ROS or ROS2 and rosbridge are accessible from the machine running the MCP client.
• The client communicates with the MCP server either via stdio or HTTP; confirm the server URL/endpoint is correctly configured in your setup (e.g., via mcp.json in the server).
• If using Gemini or other LLMs, provide API keys and ensure network access from the host running the MCP client.
• For offline or LAN-only setups, verify that all LLM providers and rosbridge endpoints are reachable without internet access.
• The project supports both ROS1 and ROS2; your rosbridge and server setup should match your ROS version.