Agent-readiness scanner

Is your API ready for AI agents?

Not an API linter — an agent-readiness check. Paste an OpenAPI spec and get a 0–100 score for how well an LLM agent can discover and call it.

or try:
Live · real scan of DigitalOcean
94/ 100
Agent-ready
635 operations scanned
  • Tool identity
  • Param clarity
  • Responses
  • Errors
  • Security
  • Ergonomics

Real scores from real specs

Live scans of well-known public APIs — these are the actual numbers.

  • 1DigitalOcean94
  • 2GitHub86
  • 3OpenAI83
  • 4Twilio Messaging80
  • 5Swagger Petstore78
  • 6Stripe69

What we score

Six categories an agent depends on, weighted by how much they affect tool-calling.

Tool identity

/25

Unique operationIds and a clear summary so an agent knows what each tool is and when to call it.

Parameter clarity

/20

Described, typed parameters — enums where bounded, required flags set — so agents fill them correctly.

Response contract

/20

Typed success responses with examples an agent can actually parse, not a bare 200 OK.

Error handling

/10

Documented 4xx errors with schemas, so an agent can recognise and recover from failures.

Security & risk

/10

Declared auth schemes and destructive operations flagged as high-risk.

Agent ergonomics

/15

Pagination, tags, and a tool count that fits the agent's context window.

Findings are free. Fixes are the product.

Every scan lists what's wrong. Unlock to get the fix for each one. A real example:

Critical · freeGET /users

Operation has no operationId — agents have no stable name to call it by.

Fix · unlocked

Add a unique, verb-led operationId (e.g. listUsers, createOrder).

How it works

01

Paste a spec

Drop in your OpenAPI JSON/YAML or a URL. Nothing to install, no account.

02

Get a score

A 0–100 agent-readiness score across 6 categories, with every finding listed for free.

03

Unlock the fixes

Pay once to reveal a concrete, copy-pasteable fix for each finding.

Questions

What does “agent-friendly” mean?
How well an LLM/MCP client can discover and call your API as a set of tools — clear names, typed params, parsable responses, documented errors.
Is this just an OpenAPI linter?
No. Linters check style and validity. agentsurface scores the specific things that decide whether an AI agent can use your API, weighted for tool-calling.
Is my spec stored?
We store the scan result under a private link with a 30-day expiry so you can share it. No account, no login.
What do I get for $19?
Every finding's concrete fix, for that scan. The score and the full finding list are always free.

Score your API in seconds

Paste a spec. Free score, free findings. Fixes when you want them.

or try: