Panel: APIs for AIs
summary
MCP is exploding since last November and integrates with GraphQL but it does not replace it: teams report 70+ internal MCP servers at one company, hallucinations rise after roughly 15–25 tools in context, and full GitHub schema dumps cost ~\$130 to seed a model while a root-query/minified approach drops that to ~\$0.06. Practical strategies shown: tree-shake operations, persistent queries, a two-step full-text-tool-index then execute flow, introspect operations (not whole schema), and watch the security “rule of two” tradeoffs around trusted input, private data, and mutations.
AI-written summary · about
chapters (14)
- 0:00AI APIs panel intro
- 2:25AI replacing developer jobs?
- 6:38MCP vs GraphQL
- 9:47MCP and GraphQL history
- 11:36GraphQL tooling for AI
- 12:57Security with AI APIs
- 15:02MCP tool authorization
- 17:48LLM token limits
- 20:19Managing context in APIs
- 21:45Two-step for large schemas
- 22:28Internal vs customer MCP tools
- 25:16Introspection for LLM efficiency
- 26:40GraphQL ops for AI
- 27:01Closing remarks
related talks
21mSimplifying MCP Tool Sprawl With GraphQL
46mPanel: AI in the Loop - Left-Shifting Platform DX and Governance
30mComposable AI: Building Next-Gen AI Agents with MCP
26mScaling Your AI Agents with Open Source: API ➔ No-Code MCP & Observability wit…
45mAgents, Tools, and MCP, oh my! Next-level AI concepts for developers
26mAsk, Don't Fetch: Why GraphQL + MCP Is the Native Language of AI Agents
33mThe Model Context Protocol: Connecting AI to Everything
50mBeyond local tools: Dive into the Model Context Protocol (MCP)