Skip to main content
Technical Deep DiveDocumentation

API Documentation and Agent Readiness: Why Mintlify Scores 66 and ReadMe Doesn't

Documentation is the bridge between an API and an AI agent. When documentation is structured, current, and machine-readable, agents can self-teach your API in seconds. When it is a stale PDF or a JavaScript-rendered page that returns blank HTML, agents see nothing. Documentation tools themselves have wildly different agent readiness — and the platform you choose directly impacts your score.

AH
AgentHermes Research
April 15, 202614 min read

The Documentation Divide: Platform Scores

Documentation platforms are an interesting test case for agent readiness because they are built to make technical information accessible. You would expect them to score well. Some do. Most do not. The gap between the best and worst documentation tools is 58 points — nearly the same range as the entire 500-business scan.

Platform
Score
Tier
Key Strengths
Mintlify
66
Silver
OpenAPI rendering, structured nav, LLM-friendly
GitBook
58
Bronze
API blocks, versioned docs, search API
Docusaurus
52
Bronze
Open source, plugin ecosystem, SSG
ReadMe
47
Bronze
Try-it console, but heavy JS rendering
Notion (as docs)
22
Not Scored
Human-friendly but agent-hostile
Google Docs / PDF
8
Not Scored
No structure, no API, not crawlable

Mintlify leads at 66 Silver because it was designed with machine-readability as a first-class concern. It generates static HTML, renders OpenAPI specs inline with structured navigation, exposes endpoint metadata in crawlable format, and ships with SEO and schema markup out of the box. When an agent crawls a Mintlify-powered docs site, it finds everything it needs without executing JavaScript.

The drop-off from there is steep. GitBook and Docusaurus are solid Bronze performers because they also generate static content but lack Mintlify's OpenAPI-first approach. ReadMe falls behind because it renders content client-side — agents that do not execute JavaScript see empty pages. And Notion, increasingly used as a documentation platform, is essentially invisible to agents.

Five Things Agents Need From Your Documentation

Agents read documentation differently than developers. They do not browse — they extract. They need five specific types of information, each mapping to a scoring dimension.

Current OpenAPI spec

D2 API Quality (0.15)

Agents auto-generate client libraries from OpenAPI. Outdated specs mean broken calls.

Example requests and responses

D6 Data Quality (0.10)

Agents learn by example. A request/response pair teaches more than paragraphs of description.

Error code reference

D6 Data Quality (0.10)

Agents need to know what 422 with code "invalid_amount" means and how to fix it. Error tables are critical.

Authentication guide

D7 Security (0.12)

How to get a token, what scopes exist, where to pass the Bearer header. Agents cannot guess auth flows.

Changelog with dates

D8 Reliability (0.13)

Agents need to know if the API changed since they last cached the schema. Breaking changes without changelogs cause silent failures.

Together, these five documentation elements directly impact D2 API Quality (0.15), D6 Data Quality (0.10), D7 Security (0.12), and D8 Reliability (0.13) — a combined 50% of the Agent Readiness Score. Documentation is not a nice-to-have. It is half your score.

The Business Documentation Problem

Documentation platforms scoring 50-66 is encouraging. But those platforms are the exception. When we look at how actual businesses document their APIs, the picture collapses. Of 500 businesses scanned by AgentHermes:

72%
No OpenAPI spec at all
45%
No public API docs
31%
Stale docs (6mo+ outdated)
18%
PDF-only documentation

The issue is not that businesses lack APIs — many have internal APIs powering their mobile apps and web frontends. The issue is that those APIs are undocumented, undiscoverable, and assume a human will figure out the integration through trial and error. Agents do not figure things out through trial and error. They read structured documentation or they move on.

AgentHermes checks documentation quality through two dimensions: D2 API Quality (0.15 weight) evaluates whether an OpenAPI spec exists and whether it is current, and D6 Data Quality (0.10 weight) checks for structured error responses, example payloads, and consistent data formats. A business can have a working API but still score poorly if agents cannot understand how to use it.

Five Documentation Anti-Patterns That Kill Agent Readiness

Anti-Pattern
Score Impact
Why It Hurts
PDF-only documentation
-15 to -25 points
Agents cannot parse PDF layouts. Every endpoint description, parameter, and example is invisible.
Swagger UI with no OpenAPI file link
-8 to -12 points
Swagger UI renders the spec visually but many deployments do not expose the raw JSON/YAML. Agents need the file, not the UI.
Outdated examples with deprecated endpoints
-5 to -10 points
Agents follow examples literally. If the example calls /v1/users but the API is now /v2/users, the agent gets a 404 and moves on to a competitor.
"Contact us for API docs"
Caps at 29
If documentation requires a sales call, agents cannot discover any endpoints. D2 and D3 both score zero. The hard cap at 29 applies.
JavaScript-rendered docs with no SSR
-5 to -8 points
Client-rendered documentation pages return empty HTML to crawlers and agents. If the content only appears after JavaScript execution, agents see a blank page.

The Documentation Fix: From Zero to Agent-Readable in a Day

Fixing documentation for agent readiness follows a specific order. Each step unlocks a scoring improvement, and the first three can be done in a single afternoon.

1

Publish an OpenAPI spec at a predictable URL

Serve your spec at /openapi.json or /api-docs. This single file unlocks D2 API Quality. Tools like swagger-jsdoc or @asteasolutions/zod-to-openapi auto-generate specs from existing code. Cost: 2-4 hours. Score impact: +8 to +12 points.

2

Add example requests and responses to every endpoint

OpenAPI supports the "example" field on every parameter, request body, and response. Fill them in with realistic data. Agents use these to construct their first API call. Cost: 1-2 hours. Score impact: +3 to +5 points.

3

Document error codes in a structured table

List every error code your API returns with its HTTP status, machine-readable code, human description, and suggested fix. Publish this as both a docs page and a JSON endpoint at /api/errors. Cost: 1 hour. Score impact: +2 to +4 points.

4

Ship an llms.txt file

A simple markdown file at /llms.txt that describes your API, links to the OpenAPI spec, and provides a quick-start guide. Takes 15 minutes and directly boosts D1 Discovery. Cost: 15 minutes. Score impact: +2 to +3 points.

5

Generate an SDK from your spec

Use openapi-generator or Speakeasy to produce typed client SDKs. Publish to npm/PyPI. Agents that find a ready-made SDK skip the API-learning step entirely. Cost: 1 hour. Score impact: +2 to +3 points.

Total impact: Steps 1-5 can lift a business from zero documentation score to +17 to +27 points across D2, D6, D7, and D8. For a business currently at 30 (Not Scored), this alone pushes into Bronze. For one at 45 (Bronze), it can push into Silver. Documentation is the single highest-ROI fix for agent readiness because it impacts the most dimensions per hour of work. See also: auto-generated SDKs for the full D2 playbook.

Frequently Asked Questions

Does documentation quality really affect my Agent Readiness Score?

Yes. Documentation directly impacts D2 API Quality (0.15 weight) and D6 Data Quality (0.10 weight), which together account for 25% of your score. A published OpenAPI spec with examples and error codes can add 10-15 points. Missing documentation caps multiple dimensions.

Why does Mintlify score higher than ReadMe?

Mintlify generates static HTML with structured content that agents can crawl directly. It renders OpenAPI specs inline with clear endpoint grouping and example blocks. ReadMe relies more heavily on client-side JavaScript rendering, which means some content is invisible to agents that do not execute JS. Mintlify also ships with better default SEO and schema markup.

Is Swagger UI enough for agent readiness?

Swagger UI is a good start but not sufficient. The key is whether you also expose the raw OpenAPI spec file at a predictable URL like /openapi.json or /api-docs. Many Swagger UI deployments render the spec visually but do not link to the downloadable file. Agents need the raw spec, not the rendered UI.

What if my documentation is in Notion?

Notion-hosted documentation scores very low for agent readiness. Notion pages are rendered client-side, require authentication for private pages, have no OpenAPI spec, and lack structured API metadata. If your docs are in Notion, the fastest fix is to export them to a static documentation platform like Mintlify or Docusaurus and add an OpenAPI spec.


How does your documentation score?

Run a free Agent Readiness Scan and see exactly how your API documentation impacts your score across D2 API Quality and D6 Data Quality.


Share this article: