kumo-rfm-mcp 0.1.0rc1

Model Context Protocol server for KumoRFM

---

The KumoRFM MCP implements a Model Context Protocol (MCP) server for interacting with the Kumo machine learning platform's RFM (Relational Foundation Model) capabilities (documentation).

Installation

The KumoRFM MCP is available for Python 3.10 and above. To install, simply run

pip install kumo-rfm-mcp

Registration

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

Available tools

The KumoRFM MCP provides several categories of tools for working with relational data and machine learning models:

Table Management Tools

• add_table: Load CSV or Parquet files into the Kumo graph
• remove_table: Remove tables from the graph
• inspect_table: View table contents and schema information
• list_tables: Get a list of all tables in the current graph

Graph Management Tools

• infer_links: Automatically detect relationships between tables based on matching column names
• inspect_graph: View the current graph structure, tables, and links
• link_tables: Manually create foreign key relationships between tables
• unlink_tables: Remove existing relationships between tables

Model Tools

• finalize_graph: Create a KumoRFM model from the current graph state for inference
• validate_query: Check if a PQL (Predictive Query Language) query is syntactically correct
• predict: Generate predictions using PQL queries on the finalized model
• evaluate: Evaluate model performance using PQL queries with ground truth data

Session Management Tools

• get_session_status: Check the current state of the session, graph, and model
• clear_session: Reset the session and clear all data

Documentation Tools

• get_docs: Access documentation, guides, and examples via kumo:// URIs

Contributing

See CONTRIBUTING.md for development setup and CLAUDE_DESKTOP_SETUP.md for Claude Desktop integration.