“GraphQL, secure? What an outlandish statement.” – and that’s exactly where this talk begins. While GraphQL is often seen as flexible and developer-friendly, its open nature makes it risky to expose directly to AI. In this session, we’ll explore how a collection of curated GraphQL operations, powered by Persisted Operations and the Model Context Protocol (MCP), turns that assumption on its head. Drawing on our work at WunderGraph, we’ll show how action-oriented, job-to-be-done operations can safely connect APIs to LLMs, keeping token usage low, requests predictable, and data access under control. Join us to discuss what secure GraphQL for AI really looks like – and how to make it work in practice.