Overview
KumoRFM MCP Server is a full-featured MCP server that empowers AI assistants with KumoRFM intelligence. It enables building, managing, and visualizing graphs directly from CSV or Parquet data, and supports converting natural language prompts into Predictive Query Language (PQL) queries for seamless interaction. You can query, analyze, and evaluate KumoRFM predictions—including missing value imputation and temporal forecasting—without any training. The server provides tools for inspecting and updating graph metadata, generating Mermaid diagrams, and materializing graphs to prepare them for inference. It also exposes I/O and model-execution capabilities such as finding table files, inspecting tables, looking up table rows, and performing predict, evaluate, and explain operations. It is compatible with Python 3.10+ and can run as a standalone process or be integrated into MCP configurations, Claude Desktop workflows, LangChain adapters, and Claude code SDK setups. This makes it easy for developers and data scientists to prototype and deploy data-driven AI assistants that reason over relational multi-table data.
Features
Graph ingestion and visualization
Build, manage, and visualize graphs directly from CSV or Parquet files.
Natural language to PQL translation
Convert natural language prompts into Predictive Query Language (PQL) queries for seamless interaction.
Inspect graph metadata
Inspect the current graph schema/metadata to understand the graph state.
Update graph metadata
Partially update the current graph schema to reflect changes in data and relationships.
Mermaid diagram generation
Return the graph as a Mermaid entity-relationship diagram for visualization.
Graph materialization
Materialize the graph based on metadata to make it available for inference operations.
I/O and data access
Find table files, inspect table structures, and lookup table rows in the raw data.
Model execution
Run predictive queries and return predictions, with support for evaluation and explanation.
