KumoRFM MCP Server

What is KumoRFM MCP Server about?

KumoRFM MCP Server enables AI assistants and developer tools to interact with the pre‑trained KumoRFM relational foundation model. It turns tabular data into a heterogeneous graph, accepts Predictive Query Language (PQL) statements—generated automatically from natural language—and returns model predictions, performance metrics, or explanations.

How to use KumoRFM MCP Server?

1. Installation – Install the Python package:

   pip install kumo-rfm-mcp
   

2. MCP configuration – Add a server entry to your MCP config (e.g., Claude Desktop mcp_config.json):

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

3. Run the server – The command above launches the server which communicates over STDIO with your agent framework.
4. Invoke tools – Use the provided MCP tools (find_table_files, inspect_graph_metadata, predict, evaluate, explain, etc.) within any agentic workflow (CrewwAI, LangChain, OpenAI agents, Claude Code SDK, etc.).

Key features

• Graph construction from CSV or Parquet files, with schema inspection and visualisation (Mermaid diagram).
• Natural‑language to PQL conversion, allowing users to ask predictive questions without writing queries.
• Model operations: predict for inference, evaluate for metric calculation, and explain for model interpretability.
• Graph management tools: lookup rows, update metadata, materialise the graph for fast inference.
• Cross‑framework adapters for CrewwAI, LangChain, OpenAI agents, Claude Desktop, and Claude Code SDK.
• Zero‑training predictions on relational multi‑table data (missing‑value imputation, temporal forecasting, etc.).

Use cases

• Business intelligence – Impute missing sales records or forecast demand directly from relational tables.
• Customer support agents – Answer complex relational queries (e.g., “Which customers bought product X after month Y?”) without custom model training.
• Data preparation – Quickly visualise and inspect data relationships before downstream analytics.
• Research prototyping – Explore relational hypotheses with instant model predictions.
• Workflow automation – Embed predictive steps into multi‑agent pipelines for finance, supply‑chain, or healthcare scenarios.

FAQ

• Do I need a GPU? No, the server runs the hosted KumoRFM model via API; only a network connection and the API key are required.
• Which Python versions are supported? Python 3.10 and newer.
• Can I run the server locally without the bundle? Yes, install the package via pip and start it with the provided command.
• What file formats are accepted? CSV and Parquet table files.
• How is authentication handled? Set KUMO_API_KEY in the environment; if omitted, an interactive OAuth2 flow is triggered on first request.

🔬 MCP server to query KumoRFM in your agentic flows

📖 Introduction

KumoRFM is a pre-trained Relational Foundation Model (RFM) that generates training-free predictions on any relational multi-table data by interpreting the data as a (temporal) heterogeneous graph. It can be queried via the Predictive Query Language (PQL).

This repository hosts a full-featured MCP (Model Context Protocol) server that empowers AI assistants with KumoRFM intelligence. This server enables:

• 🕸️ Build, manage, and visualize graphs directly from CSV or Parquet files
• 💬 Convert natural language into PQL queries for seamless interaction
• 🤖 Query, analyze, and evaluate predictions from KumoRFM (missing value imputation, temporal forecasting, etc) all without any training required

🚀 Installation

🐍 Traditional MCP Server

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

pip install kumo-rfm-mcp

Add to your MCP configuration file (e.g., Claude Desktop's mcp_config.json):

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

⚡ MCP Bundle

We provide a single-click installation via our MCP Bundle (MCPB) (e.g., for integration into Claude Desktop):

1. Download the dxt file from here
2. Double click to install

The MCP Bundle supports Linux, macOS and Windows, but requires a Python executable to be found in order to create a separate new virtual environment.

🎬 Claude Desktop Demo

See here for the transcript.

https://github.com/user-attachments/assets/56192b0b-d9df-425f-9c10-8517c754420f

🔬 Agentic Workflows

You can use the KumoRFM MCP directly in your agentic workflows:

Browse our examples to get started with agentic workflows powered by KumoRFM.

📚 Available Tools

I/O Operations

• 🔍 find_table_files - Searching for tabular files: Find all table-like files (e.g., CSV, Parquet) in a directory.
• 🧐 inspect_table_files - Analyzing table structure: Inspect the first rows of table-like files.

Graph Management

• 🗂️ inspect_graph_metadata - Reviewing graph schema: Inspect the current graph metadata.
• 🔄 update_graph_metadata - Updating graph schema: Partially update the current graph metadata.
• 🖼️ get_mermaid - Creating graph diagram: Return the graph as a Mermaid entity relationship diagram.
• 🕸️ materialize_graph - Assembling graph: Materialize the graph based on the current state of the graph metadata to make it available for inference operations.
• 📂 lookup_table_rows - Retrieving table entries: Lookup rows in the raw data frame of a table for a list of primary keys.

Model Execution

• 🤖 predict - Running predictive query: Execute a predictive query and return model predictions.
• 📊 evaluate - Evaluating predictive query: Evaluate a predictive query and return performance metrics which compares predictions against known ground-truth labels from historical examples.
• 🧠 explain - Explaining prediction: Execute a predictive query and explain the model prediction.

🔧 Configuration

Environment Variables

• KUMO_API_KEY: Authentication is needed once before predicting or evaluating with the KumoRFM model. You can generate your KumoRFM API key for free here. If not set, you can also authenticate on-the-fly in individual session via an OAuth2 flow.

We love your feedback! :heart:

As you work with KumoRFM, if you encounter any problems or things that are confusing or don't work quite right, please open a new :octocat:issue. You can also submit general feedback and suggestions here. Join our Slack!