Ask, Don't Fetch: Why GraphQL + MCP Is the Native Language of AI Agents
summary
Claims GraphQL + MCP cuts agent payloads and errors: demo shows REST returning 462 tokens vs GraphQL 18 tokens — about a 26x token reduction — and cites 73% of REST data unused and 92% of LLM API errors traced to ambiguous contracts. The team converts REST and Postgres sources into a single GraphQL+MCP endpoint with API Connect / Stepson, backing it with Postgres and testing via IBM Bob; the pipeline deploys in four commands to deploy, yields a reported ~70% backend transfer reduction, and the demo queries succeed.
AI-written summary · about
chapters (8)
- 0:00From endpoints to AI agents
- 2:01REST APIs fail for AI agents
- 4:50GraphQL solves agent challenges
- 7:01Security and schema pruning
- 8:25Transforming REST to GraphQL with MCP
- 12:01Demo: REST and PostgreSQL
- 18:42MCP integration efficiency
- 20:07Q&A
related talks
23mMCP: Is GraphQL the Right “Tool” For the Job?
27mPanel: APIs for AIs
21mSimplifying MCP Tool Sprawl With GraphQL
47mSchema Design in the Age of AI: Pragmatic Patterns to Evolve and Ship Fast
29mBuilding MCP Apps With GraphQL Patterns You Already Know
36mPutting AI Agents to Work: A New Era of Open Source Connectivity with MCP
31mWhy AI needs GraphQL
14mKeynote: GraphQL in the AI Era