Model Context Protocol · no key · read-only

Ask whether a site works for AI agents, from inside your editor.

AgentSpeed speaks the Model Context Protocol. Point your assistant at one URL and it can score a site against the public rubric, read robots.txt the way a named AI crawler would, validate structured data, and see what each crawler actually receives. The same numbers the website shows, from the same scanner.

Connect it

Claude Codeone command

claude mcp add --transport http agentspeed https://agentspeed.com/api/mcp

Claude Desktop, Cursor, and anything else that speaks MCPadd to your MCP config file

{
  "mcpServers": {
    "agentspeed": {
      "type": "http",
      "url": "https://agentspeed.com/api/mcp"
    }
  }
}

Or just call itno client required; it is plain JSON-RPC over HTTP

curl -s https://agentspeed.com/api/mcp \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Tools (8)

get_agent_readiness_scoreAgent-readiness score

Score how readable and usable a website is for AI agents. Returns a 0-100 composite, a letter grade, every check with its status, and the rubric version the score was computed under. Scores are only comparable within one rubric version.

compare_agent_readinessCompare two sites

Compare the agent-readiness of two domains: composite scores, which site wins each category, and the top findings holding the weaker one back. Both sides must already have a public scan; this tool never triggers one, so it is cheap and never spends scan budget.

can_agents_buyCan an agent buy from this store

Answer whether an AI shopping agent could complete a purchase on a store, as three ordered gates: can it FIND the products, READ the prices, and reach the CHECKOUT. Returns the first gate that is not passing, which is where the merchant should start.

get_agent_readiness_rubricThe scoring rubric itself

Return the rubric every score is computed against: version, grade bands, sub-score weights, and every check with its category, weight, and determinism class. Use this to interpret a score, or to know what a site is being graded on before changing it.

check_robots_txt_for_ai_agentsrobots.txt AI-agent verdicts

Fetch a site's robots.txt and report, per named AI agent, whether it is allowed or blocked and which rule decided it. Also returns declared sitemaps and any Content-Signal directives.

analyze_llms_txtllms.txt analysis

Fetch a site's /llms.txt, score how useful it is to an agent, and list its sections.

validate_structured_dataJSON-LD validation

Extract every JSON-LD block from a page and validate each against schema.org. Reports detected types, required-field errors, recommended-field warnings, and coverage gaps.

test_ai_crawler_accessAI crawler access matrix

Fetch a page as each of several named AI crawlers and report what each actually receives: allowed, blocked, challenged, redirected, or served different content than a browser gets.

What it will and will not do

No key, no signup. Every tool is a read-only inspection of a public website, which is the same thing the free scan page does, so there is nothing to authenticate.
Public data only. It cannot read your AgentSpeed account, your runs, or anyone’s private data. There is nothing behind it to leak.
60 calls per hour, per IP. Tool calls make real outbound requests to the site being inspected. The limit is generous for a person driving an editor and tight enough that a runaway loop stops quickly.
It honours opt-outs. A domain whose owner has asked us not to publish reports returns a refusal here exactly as it does on the website. See /bot.
Scores are version-bound. Every score carries the rubric version that produced it, and scores are only comparable within one version.

Protocol version 2025-06-18, Streamable HTTP transport. The machine-readable server card lives at /.well-known/mcp/server-card.json.