Discry Score
The Discry Score is a behavioral measurement of an API's agent-readiness, taken from its public documentation surface and always decomposed into discovery and comprehension. Discovery grades whether an agent can find a machine-legible version of the API at all; comprehension grades whether models can operate the API from the fetched docs, with answers graded mechanically against citation-verified ground truth.
The decomposition is the point. Agent-facing failures happen in order — an agent must find a usable version of the docs before its ability to understand them is ever tested — so a bare composite would hide where an API actually loses. The split shows it: a strong discovery surface routing to documentation models cannot operate from is a different problem, with a different fix, than documentation that is invisible on a plain fetch.
The score is built to be independent of who is being graded. The task banks are versioned and published, grading is mechanical rather than human judgment, every profile carries its receipts — the URLs fetched, the tasks that failed — and a contamination guard removes anything a model already knew without the docs, so fame earns nothing. The same instrument grades every API, and scoring is identical regardless of payment.
A score is produced by fetching the docs surface over plain HTTP with no JavaScript execution, quizzing models on those docs from published task banks — factual questions, request construction, and trap tasks — and grading every answer mechanically against ground truth that must cite a specific location in the API's own documentation. A separate closed-book pass establishes what models already knew, and only genuinely doc-dependent tasks count. The result publishes as discovery plus comprehension, with coverage and receipts alongside it.
What does an AI agent make of your API?
Find out in about a minute — no signup.
Discry your API — free