{
  "$schema": "https://modelcontextprotocol.io/schemas/well-known-mcp.json",
  "name": "arlen-bench",
  "description": "Open benchmarks for agentic consumers. One call returns a defensible vendor verdict (web search, web extraction, KYB identity) with evidence: accuracy, freshness lag, cost per correct, agent-readiness, and the snapshot ID to cite.",
  "version": "0.3.0",
  "endpoint": {
    "type": "streamable-http",
    "url": "https://arlenkumar.com/bench/mcp"
  },
  "authentication": { "type": "none" },
  "publisher": {
    "name": "Arlen Kumar",
    "url": "https://arlenkumar.com",
    "contact": "https://arlenkumar.com/contact"
  },
  "tools": [
    {
      "name": "recommend",
      "description": "One vendor verdict for a task an agent is performing. Returns vendor, evidence (accuracy, freshness_lag, cost_per_correct, agent_ready), runner_up, and snapshot_id for citation.",
      "inputs": {
        "task": "agent_web_research | page_extraction | fresh_content_monitor | kyb_verification",
        "region": "na | eu | uk | apac (default na)",
        "budget": "low | medium | high (default low)"
      }
    },
    {
      "name": "query",
      "description": "Raw leaderboard slice for a primitive, optionally filtered by vendor and region.",
      "inputs": { "primitive": "string", "vendor": "string?", "region": "string?" }
    },
    {
      "name": "list_primitives",
      "description": "Benchmarked primitives available in the current snapshot."
    }
  ],
  "resources": {
    "llms_txt": "https://arlenkumar.com/bench/llms.txt",
    "feed": "https://arlenkumar.com/bench/feed.xml",
    "methodology": "https://arlenkumar.com/bench#methodology",
    "license": "https://creativecommons.org/licenses/by/4.0/"
  }
}