For agents and developers

API and MCP server

Everything on the report pages, as JSON. Read-only, no key, no rate card, CORS open. The same data is exposed as tools by a hosted MCP server, so an agent can ask for a score instead of scraping for it.

REST API

Base URL https://agentfriendlyrank.com/api/v1. Responses are JSON; errors are RFC 9457 problem details with a detail that says what to do. Report and leaderboard responses are cached for five minutes, the same window as the HTML pages.

Endpoint	Returns
`GET /`	Index: links to everything below.
`GET /rubric`	The published rubric, v1.0.1: pillars, checks, weights, grade bands, changelog.
`GET /leaderboard?category=&limit=&offset=`	Every published site ranked by score. Category is one of ecommerce, saas, content, other.
`GET /sites/{domain}`	Full report for a domain: score, grade, pillars, ranked fixes, all 27 checks with evidence.
`GET /compare/{a}/{b}`	Two sites side by side, check by check. Order does not matter.
`GET /scans/{id}`	Progress of a scan in flight, for polling. Never cached.
`GET /openapi.json`	This API as an OpenAPI 3.1 document.

The full contract, with schemas, is the OpenAPI document at /api/v1/openapi.json.

Example

curl https://agentfriendlyrank.com/api/v1/sites/agentfriendlyrank.com

{
  "domain": "agentfriendlyrank.com",
  "reportUrl": "https://agentfriendlyrank.com/site/agentfriendlyrank.com",
  "scan": { "scoreTotal": 100, "grade": "A+", "pillars": { "A": 25, "B": 25, "C": 25, "D": 25 }, … },
  "topFixes": [],
  "checks": [ { "id": "A1", "status": "pass", "evidence": { … } }, … ]
}

Getting a site scanned

The API is read-only on purpose: a scan crawls someone else’s site, so requesting one stays a deliberate act with a rate limit behind it. Submit the form at /scan (it works without JavaScript, so an agent can drive it) and you are redirected to /scan/{id}. Poll /api/v1/scans/{id} until status is complete; results are cached for seven days per domain.

MCP server

A hosted Model Context Protocol server at https://agentfriendlyrank.com/mcp exposes the same data as tools. Streamable HTTP transport, stateless, no authentication: POST JSON-RPC and get JSON back. Add it to a client with:

{
  "mcpServers": {
    "agentfriendlyrank": { "url": "https://agentfriendlyrank.com/mcp" }
  }
}

Tool	Arguments
`get_site_report`	`domain`
`get_leaderboard`	`category?, limit?, offset?`
`compare_sites`	`a, b`
`get_scan_status`	`scan_id`
`get_rubric`	none

Or by hand:

curl https://agentfriendlyrank.com/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"get_site_report","arguments":{"domain":"agentfriendlyrank.com"}}}'

Terms

Free, for any use that respects the terms: cite the report URL when you republish a score, and do not present a score as current without its scan date. Scores are measured, not opinions — the methodology explains exactly how — and a hidden or robots.txt-opted-out site is absent from every endpoint, not just the HTML.

Machine surfaces

Something off?

A wrong score is a bug in a published rubric, and we treat it as one. Write to hello@agentfriendlyrank.com.