Model Context Protocol (MCP) Server

The Model Context Protocol (MCP) Server is a preview feature of the Robot Developer Extension for ROS 2. It allows Large Language Models (LLMs) to introspect a running ROS 2 system, providing a way to query and interact with the ROS 2 environment programmatically.

Model Context Protocol (MCP) is a protocol designed by Anthropic to enable LLMs to understand and interact with complex systems. It provides a structured way for LLMs to query system state, access information about components, and perform actions based on the system's context.

NOTE: This feature is in preview and may not be fully stable. It is intended for experimental use and feedback. The extension creates a python virtual environment which resides inside the extension install to run the MCP server, which is separate from the ROS 2 environment. This allows it to run independently of the ROS 2 nodes and provides a clean interface for LLMs to interact with the ROS 2 system. This is experimental code, and may cause issues with your ROS 2 system when launched. Please use with caution and report any issues you encounter.

Features

• Provides a server that can be queried by LLMs to understand the current state of a ROS 2 system.
• Allows LLMs to access information about nodes, topics, services, and parameters in the ROS 2 environment.
• Facilitates the development of intelligent applications that can leverage the capabilities of ROS 2 through natural language queries.
• Enables advanced use cases such as automated debugging, system monitoring, and dynamic configuration of ROS 2 nodes.

Getting Started

To use the MCP Server for ROS 2, Start the ROS 2 MCP Server using the command palette (CTRL-SHIFT-P) and selecting ROS2: Start MCP Server.

The first time you run the MCP Server, the extension will create a Python virtual environment. This process may require a super user password to install the required dependencie. Please open the terminal pane and enter your password in the MCP terminal if prompted.

This extension will start the server and make it available for LLMs to connect through a configurable endpoint that defaults to http://localhost:3002/sse.

Available Tools

Tool Name	Category	Description
get_nodes	Node Management	Returns a list of running ROS nodes
get_node_info	Node Management	Returns detailed information about a specific ROS node by name
list_topics	Topic Operations	Lists available ROS topics with optional type information
get_topic_info	Topic Operations	Get detailed information about a specific ROS topic
echo_topic	Topic Operations	Echo messages from a topic for monitoring
publish_to_topic	Topic Operations	Publish a message to a specific topic
list_services	Service Operations	Lists available ROS services with optional type information
get_service_type	Service Operations	Get the type of a specific ROS service
call_service	Service Operations	Call a ROS service with specified request parameters
list_parameters	Parameter Management	List all parameters of a specific node
get_parameter	Parameter Management	Get the value of a specific parameter from a node
set_parameter	Parameter Management	Set the value of a parameter for a node
list_actions	Action Operations	Lists available ROS actions with optional type information
get_action_info	Action Operations	Get detailed information about a specific ROS action
send_action_goal	Action Operations	Send a goal to an action server
record_bag	Bag Operations	Record ROS topics to a bag file
play_bag	Bag Operations	Play back a ROS bag file
get_bag_info	Bag Operations	Get information about a bag file
list_interfaces	Interface Inspection	List available ROS interfaces (messages, services, actions)
show_interface	Interface Inspection	Show the definition of a specific interface
list_packages	Package Management	List all available ROS packages
package_executables	Package Management	List executables in a specific ROS package
get_package_manifest	Package Management	Get package manifest (package.xml) information
list_launch_files	Launch Operations	List available launch files in a ROS package
launch_file	Launch Operations	Launch a ROS package with specified launch file
get_launch_parameters	Launch Operations	Get parameters available in a launch file
run_package_executable	Execution	Run a ROS package executable with optional arguments
list_lifecycle_nodes	Lifecycle Management	List nodes with lifecycle support
get_lifecycle_state	Lifecycle Management	Get the current lifecycle state of a node
set_lifecycle_transition	Lifecycle Management	Trigger a lifecycle state transition for a node
kill_process	Utility	Kill a ROS process by process ID
run_doctor	Diagnostics	Run ROS 2 doctor to check system and ROS 2 setup

Example Session (Graphical)

Example Session (Textual)

This is an example of how the MCP Server can be used to introspect a ROS 2 system:

I'll help you check the running lifecycle nodes in your ROS 2 system. Let me start by examining what lifecycle nodes are currently running.

ROS 2 (MCP Server) Now let me get information about the lifecycle nodes specifically:

ROS 2 (MCP Server) Let me get detailed information about the lifecycle node and its current state:

ROS 2 (MCP Server) Let me try without the leading slash:

ROS 2 (MCP Server) Now let me get the current lifecycle state of the lc_talker node:

ROS 2 (MCP Server) Let me also check if there are any other nodes that might be lifecycle nodes by checking the lc_listener:

ROS 2 (MCP Server) Let me try to get the lifecycle state of lc_listener as well:

ROS 2 (MCP Server) Let me get more details about the lifecycle services available for the lc_talker:

ROS 2 (MCP Server) ROS 2 (MCP Server) Based on my analysis of the running lifecycle nodes in your ROS 2 system, here's what I found:

Running Lifecycle Nodes lc_talker - Primary Lifecycle Node Current State: unconfigured (ID: 1) Node Type: This is the LifecycleTalker node from your lifecycle_talker.cpp file Purpose: A demonstration lifecycle node that publishes string messages to the /lifecycle_chatter topic

Available Lifecycle States The lc_talker node supports the standard ROS 2 lifecycle states:

Primary States:

• unknown (0)
• unconfigured (1) ← Current state
• inactive (2)
• active (3)

• finalized (4) Transition States:

• configuring (10)

• cleaningup (11)
• shuttingdown (12)
• activating (13)
• deactivating (14)
• errorprocessing (15)

Current Available Transitions From the unconfigured state, the node can transition to:

• configure → moves to configuring state (then to inactive)
• shutdown → moves to shuttingdown state (then to finalized)

Node Features The lc_talker node provides these lifecycle-specific services:

• /lc_talker/change_state - Change the lifecycle state
• /lc_talker/get_state - Get current state
• /lc_talker/get_available_states - List all possible states
• /lc_talker/get_available_transitions - Get valid transitions from current state

Topics * Publishers: /lifecycle_chatter (std_msgs/String) - publishes messages when active * System Topics: /lc_talker/transition_event - publishes state transition events

Conclusion

The included model context protocol server offers a powerful way to introspect and interact with ROS 2 systems. It allows you to query the state of nodes, topics, services, and parameters, enabling advanced use cases such as automated debugging, system monitoring, and dynamic configuration of ROS 2 nodes. We'd love to hear your feedback on this feature and any suggestions for improvements.