Stripe's OpenAPI spec has zero examples across 587 operations. None of its operations document error responses. Yet Stripe's docs are the reference founders cite for what good developer documentation looks like. The disconnect matters because agent frameworks (LangChain, OpenAI function calling, MCP) load the OpenAPI spec at runtime to build the agent's toolset. The spec is what the model sees. Not the docs site. I built AgenticScore to measure how much context each spec actually gives the agent. Six dimensions: examples, semantic clarity, error handling, intent signals, parameter documentation, pagination. Each rule weighted by how unrecoverable its absence is when an agent is constructing a call. I ran it against seven well-known APIs. Stripe scored 37 (F). OpenAI scored 32 (F). The highest score in the set was Plaid at 63 (C). No A's, no B's. The CLI is free: npx agenticscore score ./openapi.yaml Full methodology, including every weight and the reasoning: https://agenticscore.dev/methodo... Per-API teardowns: https://agenticscore.dev/leaderb... Run it on your own spec and let me know what surprises you.

AgenticScore

Score your OpenAPI spec for AI agent readiness

Score your OpenAPI spec for AI agent readiness