MCP RAG service

Open-FDD can run an optional MCP-style retrieval service derived from canonical docs.

Canonical vs derived context

  • Canonical human docs: docs/ and generated pdf/open-fdd-docs.txt.
  • Derived AI index: stack/mcp-rag/index/rag_index.json.

Never edit index artifacts as source-of-truth documentation.

Bootstrap

Run:

./scripts/bootstrap.sh --with-mcp-rag

This flow builds docs text when needed, builds retrieval index, and starts the MCP RAG service profile.

For module-focused operations, combine with bootstrap mode:

./scripts/bootstrap.sh --mode model --with-mcp-rag

Service endpoints

  • GET /health
  • GET /manifest
  • POST /tools/search_docs
  • POST /tools/get_doc_section
  • POST /tools/search_api_capabilities
  • POST /tools/get_operator_playbook

Optional guarded action tools are present but disabled by default via OFDD_MCP_ENABLE_ACTION_TOOLS=false.