Knoxville AI

Connecting a client

Install the connector in Claude, ChatGPT, Cursor, VS Code, or any MCP-capable client.

On this page

The connector URL is the same everywhere:

https://drivethruagents.com/api/mcp

Add it as a remote / custom MCP server. Below are the specifics per client.

Claude

Claude supports remote MCP servers with OAuth.

  1. Open Settings → Connectors (on claude.ai) or **Settings → Connectors → Add custom connector** (desktop app).
  2. Choose to add a custom connector and paste the URL above.
  3. The first time Claude calls a tool that needs identity, it will prompt you to sign in. Approve, and the connection is complete.

You can start using discovery immediately, before signing in.

ChatGPT

Add the URL as a custom connector / MCP server in your workspace's connector settings. ChatGPT does not currently offer a one-click install for arbitrary remote servers, so this is a paste-the-URL flow. If your client build doesn't support the OAuth handshake, use a personal API key instead.

Cursor, VS Code, and Windsurf

These support deep links that register an MCP server in a single click. The platform generates them for you: any drive thru's connector page carries a per-client install button that hands the editor the right configuration.

To wire one up by hand, add an HTTP MCP server pointing at the URL above. The configuration shape is the standard one:

{
  "mcpServers": {
    "knoxville-ai": {
      "url": "https://drivethruagents.com/api/mcp"
    }
  }
}

Any other MCP client

Anything that speaks MCP over Streamable HTTP works. Point it at the endpoint and, if it supports OAuth, let it discover the authorization server from the 401 challenge. If it doesn't, send a personal API key:

Authorization: Bearer knox_…

Scripts and CLIs

For a bare integration with no MCP client at all, the endpoint is plain JSON-RPC over HTTPS. Listing tools:

curl -s https://drivethruagents.com/api/mcp \
  -H 'Content-Type: application/json' \
  -H "Authorization: Bearer $KNOX_API_KEY" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Calling one:

curl -s https://drivethruagents.com/api/mcp \
  -H 'Content-Type: application/json' \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/call",
    "params": {
      "name": "search_drive_throughs",
      "arguments": { "query": "custom apparel", "limit": 5 }
    }
  }'

Note the second call sends no Authorization header at all — that's the anonymous path, and discovery works on it.

QR codes and text messages

Not every entry point is a connector. Participating businesses can put a QR sticker in the physical world; scanning it starts a text-message conversation with that drive thru's agent, and the message includes a link to a landing page that installs the connector in the reader's LLM app of choice.

If a multimodal assistant reads such a QR code from a photo, it can pass the decoded text straight to resolve_qr_code to find out which drive thru it points at, then open a conversation. That path is fully public.

Verifying the connection

Ask your assistant to search the directory:

Search the Knoxville drive thru directory for anything in logistics.

If you get listings back, you're connected. If your client reports an auth error on a tool like list_my_agents, that's expected until you sign in — see Authentication.

Connecting a client · Knoxville AI docs