Method

How a site gets a grade.

AgentBoard crawls public URLs the way an agent would: well-known files, OpenAPI, MCP metadata, pricing objects. No login. No JavaScript execution. Rank on the board is this score. Money cannot buy a higher grade.

  1. 1. HTTPS · 5 pts

    The site is served over HTTPS.

    Serve the site on HTTPS and redirect HTTP to HTTPS.

  2. 2. llms.txt · 12 pts

    A /llms.txt file that tells agents what you are and how to use you.

    Publish https://YOURDOMAIN/llms.txt with a short description, key URLs, and how an agent should call you.

  3. 3. Agent manifest · 15 pts

    A machine-readable agent file at /.well-known/ai-agent.json (name + capabilities).

    Publish https://YOURDOMAIN/.well-known/ai-agent.json with name, description, capabilities, contact, auth, and pricing.

  4. 4. A2A agent card · 12 pts

    An A2A agent card at /.well-known/agent-card.json.

    Publish https://YOURDOMAIN/.well-known/agent-card.json with name, url, and skills or capabilities.

  5. 5. MCP endpoint · 12 pts

    A discoverable Model Context Protocol server.

    Publish /.well-known/mcp.json (or an MCP resource metadata document) pointing at your MCP server.

  6. 6. OpenAPI spec · 12 pts

    An auto-discoverable OpenAPI or Swagger document.

    Serve /openapi.json (or /swagger.json) with a valid OpenAPI spec, or link it from the homepage.

  7. 7. Named capabilities · 8 pts

    Named tools, skills, or API operations with descriptions.

    List named capabilities in the agent manifest, agent card skills, or OpenAPI path summaries.

  8. 8. I/O schemas · 8 pts

    JSON Schema or OpenAPI schemas for inputs and outputs.

    Add components.schemas / requestBody schemas in OpenAPI, or input/output schemas on each skill.

  9. 9. Machine-readable pricing · 8 pts

    A pricing object an agent can parse — not a marketing page.

    Add a pricing object to ai-agent.json, or publish /pricing.json with currency and product amounts.

  10. 10. Auth, contact, rate limits · 8 pts

    Declared contact, authentication, and rate limits.

    Declare contact, auth, and rateLimit in ai-agent.json (and OpenAPI info.contact / securitySchemes).

User-Agent: AgentBoardBot/1.0. We refuse private IPs and metadata hosts. Scores are automated and can be wrong. Agent API.