MCP Server | AIEndpoint Docs

How it works

The MCP server gives your AI agent a 3-step fallback to understand any website:

Your agent visits site.com
        │
        ▼
  ① Check site.com/ai directly
        │
   found? ──→ use it (source: "direct", ~800 tokens)
        │
       no
        ▼
  ② Check AIEndpoint registry for cached spec
        │
   found? ──→ use it (source: "registry")
        │
       no
        ▼
  ③ Analyze the site → generate /ai spec → save to registry
        │
        ▼
  Next agent gets a cache hit at step ②

Step ① costs nothing extra — a single HTTP fetch. Step ② is a registry lookup — fast and free. Step ③ uses your agent's own context to generate the spec. The generated spec is contributed back to the community registry so every future agent benefits.

This means the MCP is useful on every website, not just those that have implemented /ai yet.

Install

Run directly with your preferred package manager — no global install needed.

npx -y @aiendpoint/mcp-server

Or install globally so it starts without a download:

npm install -g @aiendpoint/mcp-server

Then use aiendpoint-mcp as the command in your config.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "aiendpoint": {
      "command": "npx",
      "args": ["-y", "@aiendpoint/mcp-server"]
    }
  }
}

Restart Claude Desktop. You can now ask:

"Find me a free weather API" → calls aiendpoint_search_services "What can stripe.com do?" → calls aiendpoint_fetch_ai_spec

Cursor

Add to ~/.cursor/mcp.json (global) or .cursor/mcp.json (per project):

{
  "mcpServers": {
    "aiendpoint": {
      "command": "npx",
      "args": ["-y", "@aiendpoint/mcp-server"]
    }
  }
}

Claude Code

Add globally so all projects can use it:

claude mcp add aiendpoint -- npx -y @aiendpoint/mcp-server

Or scope to a single project:

claude mcp add --scope project aiendpoint -- npx -y @aiendpoint/mcp-server

Available tools

Once installed, your agent has access to these tools:

Tool	What it does
`aiendpoint_discover`	Auto-discover any website. Tries `/ai` → registry cache → auto-generate. One tool for everything.
`aiendpoint_search_services`	Search the registry by keyword, category, or auth type. Returns matching services with scores.
`aiendpoint_fetch_ai_spec`	Fetch a specific site's `/ai` endpoint directly.
`aiendpoint_validate_service`	Validate whether a site correctly implements the `/ai` spec. Returns a 0–100 score.

Example prompts

After setting up the MCP server, try asking your agent:

"What can github.com do?" → aiendpoint_discover checks /ai, falls back to registry or generates a spec
"Find me a free weather API" → aiendpoint_search_services queries the registry
"Does stripe.com have a valid /ai endpoint?" → aiendpoint_validate_service runs the checker