The State of API Documentation

About this paper

API documentation is the software companies buy so that their APIs can be understood. It sells rendering, hosting, versioning, search, design, and — increasingly — the machine-readable surface that lets an AI agent read a developer portal instead of a human. It is the market that produces legibility as a product. That makes the question “is this vendor’s own surface legible?” a product demonstration rather than a rhetorical flourish.

Documentation is also a fuzzy category, and the tag roll-up is not a reliable boundary — it reports 87 providers for this area, a looser and different set than the market. So this cohort is 29 named by hand from the curated area index: the 15 hosted platforms you buy, the 8 reference renderers you embed, and the 6 docs-as-code generators you build with. SDK generators, API clients, linting tooling and internal developer portals are real adjacent markets, each covered in its own area rather than folded in here.

This market is genuinely strong, and that should be said first. It averages 43.3 against a whole-catalog average of 22.8, puts 55.2% of its cohort in the upper three bands against 19.5% catalog-wide, and lands not one provider in the Minimal band where 42.2% of the catalog sits. Every provider in it can be found and states clearly what it does.

Its weakest facet is developer ergonomics, at 25.6 — the lowest of the six, in the market that sells developer experience. Two cheap unfilled fields drive most of it: examples at 17.2% and error semantics at 17.2%, inside specifications these vendors already publish. That is the market’s largest single opportunity, and the report treats it as a roadmap rather than a verdict.

And then four clean zeros. An MCP server 0 of 29. Idempotency 0 of 29. Consent and delegated identity 0 of 29. Dry-run mode 0 of 29. Four vendors in this cohort actively sell MCP servers and AI search to their customers, and none of the twenty-nine runs one on its own documentation. Scopes come back at zero across all twenty-nine, in a market whose permission model — read, draft, publish, administer — is already built into every one of these products for humans.

This is the second Area Market Report — a cut of the catalog by what an API does rather than who it sells to. The full report maps the three-tier market structure, the four frontier resources in the entire cohort, the facet and dimension data behind every number above, the standards layer where this market authored the specification it renders, and the agent-readiness ranking that reorders the market completely. It closes on a provenance audit of a score with two authors: which numbers here are the market’s, which are the catalog’s, and why the zeros are the ones that carry the report.

What's inside

1. Executive summary — the market that sells legibility
2. The state of the market — three products, three buyers, three prices
3. The resource taxonomy — what a documentation API exposes
4. The composite Kin Score — and the ergonomics finding
5. Agent readiness — and four clean zeros
6. The blueprint — what the leaders satisfy, and what nobody does
7. Provenance — a score with two authors
8. The standards layer — the market that wrote the specification it renders
9. The regulatory layer — no mandate, and no permission to wait for
10. Artifacts for the agentic turn
11. Security, authentication, scopes
12. Provider-by-provider intelligence
13. The investable thesis
14. Where this is going

What you get for $500.00