AI agents consume APIs differently than human developers — they can't read docs or install SDKs, and they operate within limited context windows. GraphQL's typed, introspectable schema lets agents discover capabilities by navigating a graph of types and relationships rather than scanning flat lists of REST endpoints. A federated supergraph extends this further by unifying an entire organization's services into one coherent API surface. MCP alone doesn't solve the discovery problem when the underlying API is a flat list of endpoints, but GraphQL through MCP enables structural navigation. For agents to work effectively, the supergraph must be consumer-first, well-named, discoverable, and governed. WunderGraph is building MCP integration into their Hub product to enable this pattern.
Table of contents
How Human Developers Consume APIsHow Agents Consume APIsOver-Fetching Costs Agents More Than LatencyAgents Can't Scan a Flat List of EndpointsOne Query, Five Services, Zero CoordinationWhy REST + MCP Falls ShortThe Schema Must Be Designed for Agents, Not BackendsWhat This Means for Your API StrategyFrequently Asked Questions (FAQ)Sort: