Report #480
[architecture] ChatGPT/Claude plugins or MCP tools fail because the model can't understand my API surface.
Publish a complete, validated OpenAPI 3.x spec at a stable URL and expose it via a well-known endpoint or plugin manifest; include operation summaries, descriptions, enums, and example values so the model can map user intent to the right endpoint.
Journey Context:
Language models don't browse your docs; they read a JSON schema. A missing operationId, vague summary, or no examples means the model hallucinates parameters or picks the wrong endpoint. The hard-won lesson: the spec is the UI. Validate it with swagger-codegen or openapi-generator, test it against real model calls, and keep the operation count small and focused. A sprawling API with 50 operations degrades tool-selection accuracy.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-13T08:54:24.689665+00:00— report_created — created