API reference
An API reference is the section of documentation that describes every endpoint in contract-level detail: URL, HTTP method, parameters, request and response schemas, error codes, and examples. It is the material an integrator consults while writing actual calls, complementing guides and tutorials that explain concepts and workflows at a higher level.
AI agents read references literally. A human integrator fills gaps with intuition and trial; a model constructs its next API call from exactly what the reference states. Endpoint descriptions that explain what an operation accomplishes — rather than restating its shape — and request examples that are runnable as written are the difference between an agent that composes a correct call and one that hallucinates parameters.
Delivery format matters as much as content. Most agents fetch documentation with plain HTTP and no JavaScript execution, so a reference rendered entirely client-side arrives as an empty shell. References available as clean markdown, or mirrored through llms.txt, reach agents in a form they can actually parse.
The reference is also where execution-critical details live or die: pagination mechanics, rate limits, error formats, and idempotency behavior are useless to an agent if they exist only in tribal knowledge or a support FAQ.
The API reference is the core of the public documentation surface Discry scans — fetched with plain HTTP, no JavaScript, exactly as most agents fetch it. Comprehension checks on the profile record whether endpoints are described task-first, whether examples are realistic and runnable, whether multi-step workflows are chained into complete jobs, and whether the docs are available as markdown an agent can parse.
What does an AI agent make of your API?
Find out in about a minute — no signup.
Discry your API — free