DSR

You can use the DSR MCP Server to query, inspect, and modify Deep State Representation graphs through standardized MCP tools and real-time graph operations. This server lets you retrieve nodes and edges, insert or update elements, and save graph states, enabling AI assistants to interact with robot perception and scene understanding data efficiently.

How to use

You interact with the DSR MCP Server through an MCP client by selecting the appropriate tool to perform tasks such as retrieving nodes, filtering by type, inspecting node details, creating or deleting graph elements, and saving the graph state. Start and test connections using the provided server endpoints or local MCP commands, then use the tools to perform read and write operations as needed. The server exposes both read-only resources for efficient data access and modification tools for updating the graph.

How to install

Prerequisites: you need Python 3.12+ installed and a compatible MCP client environment. You also have options to run the server via a local Python installation or via the UV-based workflow for MCP servers.

# Option 1: Install with UV (recommended for MCP clients)

# 1) Clone the repository
git clone https://github.com/grupo-avispa/dsr_mcp_server.git

# 2) Enter the project directory
cd dsr_mcp_server

# 3) Install the UV-based runtime and dependencies
uv sync

# 4) Start the MCP server locally
uv run dsr_mcp_server

Option 2: Install with pip (Python package installation) and run with Python. The server can be started by invoking its module directly.

# 1) Clone the repository
git clone https://github.com/grupo-avispa/dsr_mcp_server.git

# 2) Enter the project directory
cd dsr_mcp_server

# 3) Install the package in editable mode
python3 -m pip install .

# 4) Start the MCP server
python3 -m dsr_mcp_server

Additional configuration and transport options

The server supports multiple transport modes. You can use the UV-based local command to run the server, or configure an HTTP transport for remote access. The following configurations illustrate two common setups.

{
  "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"
    }
  }
}

For HTTP transport, you can expose the MCP server at a URL suitable for your integration. The HTTP transport example shows how to point a client to a local HTTP endpoint.

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

Available tools

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)

get_node_details

Get detailed information about a specific node including attributes and edges

get_all_edges

Retrieve all edges from the DSR graph

insert_node

Insert a new node into the DSR graph

insert_edge

Insert a new edge between two nodes in the DSR graph

insert_edge_attribute

Insert or update an attribute for an edge in the DSR graph

update_node

Update a node with new attributes in the DSR graph

delete_node

Delete a node from the DSR graph

delete_edge

Delete an edge from the DSR graph

save_graph

Save the current state of the DSR graph to a JSON file