generated: '2026-07-19'
method: searched
source: https://github.com/kumo-ai/kumo-rfm-mcp
status: published
servers:
- name: kumo-rfm-mcp
  title: KumoRFM (Relational Foundation Model)
  description: Full-featured MCP server that lets AI assistants build relational graphs from CSV/Parquet
    files, write Predictive Query Language (PQL) queries, and run training-free predictions, evaluations,
    and explanations against KumoRFM.
  repository: https://github.com/kumo-ai/kumo-rfm-mcp
  package:
    registry: pypi
    name: kumo-rfm-mcp
    version: 0.3.1
    install: pip install kumo-rfm-mcp
  transports:
  - stdio
  - streamable-http
  command: python -m kumo_rfm_mcp.server
  http_flags: --transport streamable-http --host 0.0.0.0 --port 8000 --path /mcp
  auth:
    env: KUMO_API_KEY
    http_bearer_env: MCP_BEARER_TOKEN
    notes: 'Set KUMO_API_KEY for headless deployments to avoid the browser-based OAuth flow; clients that
      cannot inject env vars may call the `authenticate` tool with an `api_key` argument once at session
      start. When MCP_BEARER_TOKEN is set the HTTP endpoint requires an `Authorization: Bearer <token>`
      header.'
  bundle:
    format: MCPB (.dxt)
    url: https://kumo-sdk-public.s3.us-west-2.amazonaws.com/dxt/kumo-rfm-mcp-0.2.0.dxt
  install_claude_code: claude mcp add --transport stdio kumo-rfm-mcp --env KUMO_API_KEY=<YOUR-API-KEY>
    -- python -m kumo_rfm_mcp.server --port 8000
  tools:
  - name: get_docs
    description: Get documentation on how to use KumoRFM (overview, graph-setup, predictive-query, explainability).
    source: kumo_rfm_mcp/tools/docs.py
  - name: authenticate
    description: Authenticate the current KumoRFM session.
    source: kumo_rfm_mcp/tools/auth.py
  - name: find_table_files
    description: Finds all table-like files (e.g., CSV, Parquet) in a directory.
    source: kumo_rfm_mcp/tools/io.py
  - name: inspect_table_files
    description: Inspect the first rows of table-like files.
    source: kumo_rfm_mcp/tools/io.py
  - name: inspect_graph_metadata
    description: Inspect the current graph metadata.
    source: kumo_rfm_mcp/tools/graph.py
  - name: update_graph_metadata
    description: Partially update the current graph metadata.
    source: kumo_rfm_mcp/tools/graph.py
  - name: get_mermaid
    description: Return the graph as a Mermaid entity relationship diagram.
    source: kumo_rfm_mcp/tools/graph.py
  - name: materialize_graph
    description: Materialize the graph based on the current state of the graph metadata.
    source: kumo_rfm_mcp/tools/graph.py
  - name: lookup_table_rows
    description: Lookup rows in the raw data frame of a table for a list of primary keys.
    source: kumo_rfm_mcp/tools/graph.py
  - name: predict
    description: Execute a predictive query and return model predictions.
    source: kumo_rfm_mcp/tools/model.py
  - name: evaluate
    description: Evaluate a predictive query and return performance metrics.
    source: kumo_rfm_mcp/tools/model.py
  - name: explain
    description: Execute a predictive query and explain the model prediction.
    source: kumo_rfm_mcp/tools/model.py
  resources:
  - uri: kumo://docs/overview
    name: Overview of KumoRFM
  - uri: kumo://docs/graph-setup
    name: Graph Setup
  - uri: kumo://docs/predictive-query
    name: Predictive Query
  - uri: kumo://docs/explainability
    name: Explainability
- name: kumo-docs-mcp
  title: Kumo / NVIDIA SDGM documentation MCP server
  description: Hosted MCP server over the Kumo (Structured Data and Graph Models) documentation, advertised
    in the docs llms.txt for AI client integration.
  url: https://docs.nvidia.com/sdgm/_mcp/server
  transports:
  - http
  source: https://docs.nvidia.com/sdgm/llms.txt
notes: KumoRFM API keys are issued at https://kumorfm.ai.
deployment:
  mode: none
  endpoint: https://kumo-sdk-public.s3.us-west-2.amazonaws.com/dxt/kumo-rfm-mcp-0.2.0.dxt
  verified: probed
  probe: dead
  note: the endpoint this manifest claimed did not answer; recorded as none rather than deleted so the
    claim stays auditable
  checked: '2026-08-12'
  source: catalog MCP census