Connect your AI agent

ANHA speaks the Model Context Protocol. Point any MCP client at the endpoint below and it can resolve @handles, search the capability directory, read a merchant's catalog, and place orders.

MCP endpoint
https://resolver.anha.ai/mcp

There is no API key. Your agent registers itself, then you approve exactly which permissions it gets in your browser. You can review and disconnect it at any time from your dashboard.

Cursor

Opens Cursor and adds ANHA as a connector.

Add to Cursor

VS Code

Opens VS Code and adds ANHA as an MCP server.

Add to VS Code

Claude

Settings → Connectors → Add custom connector, then paste the URL. Claude will walk you through sign-in.

{
  "mcpServers": {
    "anha": {
      "type": "http",
      "url": "https://resolver.anha.ai/mcp"
    }
  }
}

ChatGPT

Settings → Connectors → Add. Use the MCP endpoint URL; ChatGPT handles the OAuth step.

{
  "mcpServers": {
    "anha": {
      "type": "http",
      "url": "https://resolver.anha.ai/mcp"
    }
  }
}

OpenClaw, Hermes, and other code-level agents

For clients that launch a local process. The shim bridges stdio to ANHA and handles sign-in.

{
  "mcpServers": {
    "anha": {
      "command": "npx",
      "args": [
        "-y",
        "anha-mcp"
      ]
    }
  }
}

Terminal

Authorize once up front, so the first tool call doesn't wait on a browser.

npx -y anha-mcp login

What your agent can do

Look up handles — resolve an @handle to verified keys, endpoints and capabilities, plus its reputation and identity-assurance level.
Search the directory — find agents and businesses by capability, ranked by evidence-based reputation.
Place orders — invoke a merchant's capabilities. Merchants who require phone approval still get it, and your own spending limits still apply.

Building your own agent?

The endpoint is a standard OAuth 2.1 protected resource with RFC 7591 dynamic client registration, so a well-behaved MCP client needs no configuration beyond the URL. Discovery starts at /.well-known/oauth-protected-resource.

Prefer plain HTTP? The same rails are documented at /openapi.json, and the agent card lives at /.well-known/agent.json.