DSR MCP Server

An MCP (Model Context Protocol) server that provides a comprehensive set of tools to query, manipulate, and analyze Deep State Representation (DSR) graphs. This server enables AI assistants to interact with robot perception and scene understanding data through standardized MCP tools with real-time graph operations and intelligent node relationship management.

Tools

Tool Name	Description	Parameters
check_dsr_connection	Check DSR connection and return status information	—
get_all_nodes	Retrieve all nodes from the DSR graph	—
get_nodes_by_type	Filter nodes by their type (e.g., robot, person, room)	node_type: str
get_node_details	Get detailed information about a specific node including attributes and edges	node_identifier: str
get_all_edges	Retrieve all edges from the DSR graph	—
insert_node	Insert a new node into the DSR graph	name: str, node_type: str
insert_edge	Insert a new edge between two nodes in the DSR graph	origin_id: str, destination_id: str, edge_type: str
insert_edge_attribute	Insert or update an attribute for an edge in the DSR graph	origin_id: str, destination_id: str, attribute_name: str, attribute_value: str, attribute_type: str = 'string'
update_node	Update a node with new attributes in the DSR graph	node_id: str, attribute_name: str, attribute_value: str, attribute_type: str = 'string'
delete_node	Delete a node from the DSR graph	node_id: str
delete_edge	Delete an edge from the DSR graph	origin_id: str, destination_id: str, edge_type: str
save_graph	Save the current state of the DSR graph to a JSON file	Interactive file path selection

Resources

Resources provide read-only, efficient access to DSR graph data. They are ideal for querying information without modifying the graph state.

Resource URI	Description	Returns
dsr://nodes	All nodes in the DSR graph with basic information	JSON with nodes array, count, and DSR name
dsr://nodes/type/{type}	Nodes filtered by type with full details (attributes, edges)	JSON with detailed nodes array, count, and type
dsr://nodes/{node_id}	Detailed information about a specific node	JSON with node details, attributes, and connected edges
dsr://edges	All edges in the DSR graph	JSON with edges array, count, and DSR name

Environment Variables

Variable	Default	Description
DSR_AGENT_ID	42	Unique agent identifier for DSR connection
DSR_NAME	mcp_server	Target DSR graph name to connect to
SERVER_HOST	127.0.0.1	Server host address
SERVER_PORT	3000	Server port number

Installation

Dependencies

• fastmcp: MCP server framework (>= 2.2.7)
• Cortex: DSR library for graph operations
• Python: 3.12+ required

Note: You must build Cortex inside the virtual environment to ensure compatibility.

cd dsr_mcp_server
source .venv/bin/activate
cd ${CORTEX_DIR} && make -p build && cd build
cmake .. && make -j$(nproc) && sudo make install

Install with uv (recommended)

Clone the repository and install with uv:

git clone https://github.com/grupo-avispa/dsr_mcp_server.git
cd dsr_mcp_server
uv sync

Or install directly from the repository:

uv add git+https://github.com/grupo-avispa/dsr_mcp_server.git

Install with pip

Install the package in mode:

git clone https://github.com/grupo-avispa/dsr_mcp_server.git
cd dsr_mcp_server
python3 -m pip install .

Or install directly from the repository:

python3 -m pip install git+https://github.com/grupo-avispa/dsr_mcp_server.git

Usage

Running with uv

uv run dsr_mcp_server

Running with pip installation

python3 -m dsr_mcp_server

The server will start and attempt to initialize the DSR connection automatically using the configured parameters.

Configuration example for Claude Desktop/Cursor/VSCode

Using uv (recommended)

Add this configuration to your application's settings (mcp.json):

{
  "dsr mcp server": {
    "type": "stdio",
    "command": "uv",
    "args": [
      "run",
      "--directory",
      "/path/to/dsr_mcp_server",
      "dsr_mcp_server"
    ],
    "env": {
      "DSR_AGENT_ID": "42",
      "DSR_NAME": "your_dsr_graph_name"
    }
  }
}

Using pip installation

{
  "dsr mcp server": {
    "type": "stdio",
    "command": "python3",
    "args": [
      "-m",
      "dsr_mcp_server"
    ],
    "env": {
      "DSR_AGENT_ID": "42", 
      "DSR_NAME": "your_dsr_graph_name"
    }
  }
}

HTTP Server Mode

For HTTP transport integration:

{
  "servers": {
    "dsr_mcp_server": {
      "type": "http",
      "url": "http://localhost:3000/mcp"
    }
  }
}

Technical Notes

• Connection to DSR is performed with automatic initialization and connection monitoring.
• Node attributes are filtered to exclude internal rendering properties (color, depth, height, level, etc.) for cleaner output.
• Edge relationships support various types including spatial (near), ownership (has), identity (is), and association (is_with) semantics.
• Graph operations maintain consistency through the DSR library's built-in validation mechanisms.
• All tools return standardized JSON responses with success/error status and detailed information.
• Interactive file selection for graph export operations through MCP elicit mechanism.
• Resources provide read-only, idempotent access to DSR graph data with efficient caching and lower overhead compared to tools. Use resources for querying data and tools for modifications.

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes following the coding conventions
4. Add tests if applicable
5. Submit a pull request

Support

For issues, questions, or contributions, please refer to the project's issue tracker or contact the development team.