The terms that decide whether you can use an API are barely documented.

We checked the access terms of 1,802 public APIs — authentication, pricing, free tiers, rate limits, specs, and MCP servers. Every value links back to the vendor's own documentation. Here's what the machine-readable side of the API economy looks like in mid-2026.

These are the questions every integration starts with — how do I auth, what does it cost, what's free, where are the limits — and for most public APIs the vendor's own pages answer only some of them. That documentation gap is the reason this census exists: we record what is stated, link the page that states it, and publish an honest null for the rest.

A note on a retired claim: an earlier version of this report led with an MCP-vs-OpenAPI ratio. We pulled it — a direct probe of live domains finds far more OpenAPI specs than our extraction captured, so the two numbers were not comparable floors. When our own data can't support a claim, the claim goes.

Free tiers are normal now

Among the APIs where we could fully verify the terms, 90% offer some kind of free tier and 57% publish their rate limits. Free access isn't unusual anymore — it's what developers expect.

The strange part: government and open-data APIs are among the least likely to clearly document a usable free tier. The APIs may be free, but whether you can rely on them in production is often simply not stated.

Authentication is simpler than it looks

For APIs that document authentication, API keys still dominate — and almost one in five APIs needs no authentication at all. OAuth gets a lot of attention in integration guides, but it's still the minority case. For agents calling lots of different APIs, that's good news: most authentication comes down to a single static credential.

Pricing has settled into a pattern

Most APIs are either freemium or completely free. Pure pay-to-play is relatively uncommon, and usage-based pricing is still a minority. The default model is pretty clear by now: give people a free bucket, then charge based on usage, users, or features.

Most APIs are still hard for machines to discover

Across the live API domains we checked, only 80% publish an llms.txt file. For most public APIs, an agent still has to read the same documentation pages a human would. That's the gap this census is trying to close.