KumoRFM

The KumoRFM MCP Server lets you query a Relational Foundation Model directly from your AI agents. It builds and serves an interface to load tabular data, translate natural language into predictive queries, and return predictions without training. You can run it locally as a stdio process or connect to it via an MCP client to integrate KumoRFM intelligence into your workflows.

How to use

You run the MCP server locally and connect your MCP client to it. The server exposes a standard set of tools to inspect graphs, manage data sources, generate predictive queries, and evaluate results. You can start by configuring your client to launch the server as a stdio process using Python, then pass your KUMO_API_KEY for authentication. Once connected, you can ask your agent to load CSV or Parquet files, build a graph from those sources, visualize the graph, and run predictive queries against KumoRFM.

How to install

Prerequisites: ensure you have Python 3.10 or newer installed on your system. You also need pip to install Python packages.

Install the MCP server package using pip and then configure the MCP server in your client setup.

pip install kumo-rfm-mcp

{
  "mcpServers": {
    "kumo-rfm": {
      "command": "python",
      "args": ["-m", "kumo_rfm_mcp.server"],
      "env": {
        "KUMO_API_KEY": "your_api_key_here"
      }
    }
  }
}

Additional notes and configuration

Set the API key you received for KumoRFM in the environment when starting the server. If you prefer not to provide it up front, you can authenticate on-the-fly during the session via an OAuth2 flow.

To include the server in an MCP-based workflow, you typically start it as a stdio process and pass the API key through the environment. The common start command is the Python module invocation shown above, wrapped in your MCP client configuration.

Available tools

find_table_files

Searches a directory to locate tabular data sources such as CSV and Parquet files.

inspect_table_files

Inspects the structure and sample rows of table-like files to understand schema and content.

inspect_graph_metadata

Shows the current graph metadata, including nodes, edges, and attributes.

update_graph_metadata

Partially updates the graph schema to add or modify metadata.

get_mermaid

Generates a Mermaid ER diagram representing the current graph structure.

materialize_graph

Materializes the graph from metadata to make it available for inference.

lookup_table_rows

Retrieves rows from a table by primary keys for quick data checks.

predict

Runs a predictive query against KumoRFM and returns model predictions.

evaluate

Evaluates a predictive query, returning performance metrics against ground-truth data.

explain

Explains the model prediction for a given query.