Discover and Verify AI Services

Not Human Search is a remote MCP server letting AI agents search a curated index of 1,750+ AI-ready websites, inspect indexed site details, submit new sites for analysis, and verify live MCP endpoints via a JSON-RPC probe - designed for agents that need to discover tools, APIs, and services at runtime rather than relying on hardcoded lists. The endpoint uses streamable HTTP and requires no authentication or API key.

It exposes seven tools: search_agents (keyword search over the index, returning ranked results with scores, categories, and available endpoints), get_site_details (a specific domain's AI-readiness score and available machine-readable endpoints), get_stats (aggregate index statistics), submit_site (submit a URL for crawling and AI-readiness analysis), verify_mcp (send a JSON-RPC probe to a URL to confirm it's a live, responding MCP endpoint), list_categories (available discovery categories), get_top_sites (top-ranked indexed sites), and register_monitor (register a domain monitor tied to a user-provided email address).

{
  "mcpServers": {
    "not-human-search": { "url": "https://nothumansearch.ai/mcp" }
  }
}

Use it when an agent needs to discover tools, APIs, or MCP servers for a specific task, check whether a website exposes machine-readable endpoints (llms.txt, OpenAPI, MCP), verify that a specific MCP endpoint is actually live, or build a workflow that dynamically finds and connects to external services. Best practice is search_agents for broad discovery followed by get_site_details for specifics, and verify_mcp to confirm liveness before wiring an endpoint into a workflow - only probe URLs you expect to be MCP endpoints, since verify_mcp sends a live JSON-RPC request. register_monitor requires a user-provided email address and explicit consent to receive monitoring notifications - only use it with an email the user has actually supplied. Scoring reflects machine-readable signals (llms.txt, OpenAPI, MCP, structured data), not content quality, and the index (1,750+ sites) doesn't cover the entire internet.

Input is a search query, a domain to inspect, a URL to submit or verify, or a domain/email pair to monitor. Output is ranked search results, a site's AI-readiness score and endpoint breakdown, index statistics, submission confirmation, or a JSON-RPC verification result confirming (or denying) that a URL is a live MCP endpoint.

Exposed as a remote, no-auth MCP server (https://nothumansearch.ai/mcp) compatible with Claude Desktop, Cursor, and Windsurf; related to companion skills @mcp-builder (building MCP servers) and @ai-dev-jobs-mcp (AI/ML job search via MCP).

Developers building agent workflows that need runtime discovery of tools, APIs, or MCP servers rather than hardcoded service lists, or that need to verify a specific MCP endpoint is live.