API RAG is a documentation assistant that lets you ask API questions in plain English and get direct answers with source citations.
It ingests an OpenAPI spec, indexes endpoint/schema chunks in Vertex AI Search, retrieves relevant context, and uses Gemini to generate grounded responses. The same retrieval pipeline is also exposed as an MCP server for assistant-driven coding workflows.
What it is
A practical API docs assistant focused on reliable answers:
- Grounded Answers: Uses retrieved documentation chunks instead of free-form guessing.
- Citations Included: Shows source snippets so answers are easy to verify.
- OpenAPI Ingestion: Supports URL or local specs and outputs JSONL chunks.
- Two Interfaces: Works as a web app and as an MCP server.
How It's Built
- Python Pipeline: Ingestion, retrieval, and generation logic.
- Vertex AI Search: Indexes endpoint and schema chunks for fast retrieval.
- Gemini Integration: Produces structured responses from retrieved context.
- Gradio UI: Chat interface with source panel and streaming answers.
- MCP Tools:
search_docsandget_schemafor assistant integrations.