MCP Server for LLM Integration

ros2_medkit_mcp provides MCP (Model Context Protocol) tools for connecting Large Language Models to your ROS 2 system via the ros2_medkit gateway.

Overview 

The MCP server enables:

Natural language queries about robot state
LLM-assisted diagnostics and troubleshooting
Autonomous agents for robot monitoring and control

The server wraps the gateway’s HTTP API, making it accessible to LLMs through standardized MCP tools.

Quick Start 

Prerequisites 

Python 3.11+
Poetry
A running ros2_medkit gateway

Installation 

git clone https://github.com/selfpatch/ros2_medkit_mcp.git
cd ros2_medkit_mcp
poetry install

Running the Server 

stdio transport (for Claude Desktop, VS Code):

poetry run ros2-medkit-mcp-stdio

HTTP transport (for remote access):

poetry run ros2-medkit-mcp-http --host 0.0.0.0 --port 8765

Configuration 

The server is configured via environment variables:

Variable	Default	Description
`ROS2_MEDKIT_BASE_URL`	`http://localhost:8080/api/v1`	Base URL of the gateway API
`ROS2_MEDKIT_BEARER_TOKEN`	(none)	Optional JWT token for authentication
`ROS2_MEDKIT_TIMEOUT_S`	`30`	HTTP request timeout in seconds

Claude Desktop Integration 

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "ros2_medkit": {
      "command": "poetry",
      "args": ["run", "ros2-medkit-mcp-stdio"],
      "cwd": "/path/to/ros2_medkit_mcp",
      "env": {
        "ROS2_MEDKIT_BASE_URL": "http://localhost:8080/api/v1"
      }
    }
  }
}

VS Code Integration 

Create .vscode/mcp.json in your workspace:

Option 1: stdio transport (local)

{
  "servers": {
    "ros2_medkit": {
      "type": "stdio",
      "command": "poetry",
      "args": ["run", "ros2-medkit-mcp-stdio"],
      "cwd": "/path/to/ros2_medkit_mcp",
      "env": {
        "ROS2_MEDKIT_BASE_URL": "http://localhost:8080/api/v1"
      }
    }
  }
}

Option 2: HTTP transport (Docker/remote)

{
  "servers": {
    "ros2_medkit": {
      "type": "sse",
      "url": "http://localhost:8765/mcp"
    }
  }
}

Docker Usage 

# Run HTTP server
docker run -p 8765:8765 \
  -e ROS2_MEDKIT_BASE_URL=http://host.docker.internal:8080/api/v1 \
  ghcr.io/selfpatch/ros2_medkit_mcp:latest

# Run with stdio transport
docker run -i ghcr.io/selfpatch/ros2_medkit_mcp:latest stdio

MCP Tools Reference 

Discovery Tools 

Tool	Description
`sovd_version`	Get SOVD API version information
`sovd_entities_list`	List all entities (areas, components, apps, functions)
`sovd_entities_get`	Get a specific entity by ID
`sovd_area_components`	List components within an area

Faults Tools 

Tool	Description
`sovd_faults_list`	List faults for a component
`sovd_faults_get`	Get specific fault details
`sovd_faults_clear`	Clear/acknowledge a fault

Data Tools 

Tool	Description
`sovd_entity_data`	Read all topic data from an entity
`sovd_entity_topic_data`	Read data from a specific topic
`sovd_publish_topic`	Publish data to a topic

Operations Tools 

Tool	Description
`sovd_list_operations`	List operations (services/actions) for a component
`sovd_create_execution`	Call a service or send an action goal
`sovd_get_execution`	Get action execution status
`sovd_list_executions`	List all executions for an operation
`sovd_cancel_execution`	Cancel a running action

Configurations Tools 

Tool	Description
`sovd_list_configurations`	List parameters for a component
`sovd_get_configuration`	Get a parameter value
`sovd_set_configuration`	Set a parameter value
`sovd_delete_configuration`	Reset parameter to default
`sovd_delete_all_configurations`	Reset all parameters

Example Prompts 

Once configured, try these prompts with your LLM:

“List all components in the system”
“What faults are reported by the lidar sensor?”
“Show me the current temperature reading”
“Call the calibrate service on the camera component”
“What parameters does the sensor node have?”

Repository 

https://github.com/selfpatch/ros2_medkit_mcp