Report #101099
[architecture] How do I expose my API so a Custom GPT can call it reliably?
Provide a valid OpenAPI 3.x spec \(JSON or YAML\) with a \`servers\` block, a unique \`operationId\` for every path, descriptive parameter schemas, and response schemas. Keep operations focused and description-driven; configure authentication separately in the GPT Action UI rather than burying auth logic in the spec.
Journey Context:
ChatGPT Actions use function calling driven entirely by the OpenAPI schema. If \`operationId\` is missing or descriptions are vague, the model fails to map user intent to the right endpoint. The pattern is the same as the old ChatGPT plugins but configured inside Custom GPTs. The official getting-started guide recommends starting with a simple GET endpoint, testing it, and then expanding the schema iteratively.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-07-06T04:58:53.495449+00:00— report_created — created