Building an MCP Server for a Component
summary
AG Grid builds an MCP server called agarch to let LLMs use versioned docs: it ingests 360,000 docs, chunks by H2, embeds each chunk with OpenAI embeddings, stores vectors in Postgres + PGVector, and exposes prompts, resources and a tool API. The tool takes a query plus framework and version, does a hybrid vector/text search, returns condensed markdown, and Claude installs the MCP client and calls a quick start prompt that generates a working React AG Grid component with sorting, filtering, row grouping and editing.
AI-written summary · about
chapters (7)
- 0:00Introduction to AG Grid
- 1:41The context problem
- 3:48Understanding MCP
- 5:32Core MCP primitives
- 7:27Building an optimized search
- 9:56Demo and conclusion
- 10:38Closing remarks
related talks
29mMy Context Progress - A Psychologists Journey to Implementing an MCP
21mBuilding Agents (the hard parts!)
17mBeyond Components: Designing Generative UI for MCP Apps
23mMCP = Mega Context Problem
49mThe Headless Web
07mQ&A with Kent C. Dodds
37mLetting AI Interface with Your App with MCPs
27mAI-Generated React Server Components