MCP setup

The Model Context Protocol lets an AI assistant call external tools. Point a client at our MCP endpoint and every endpoint in the catalogue becomes a tool the model can invoke on its own - with the credit cost included in each tool description so it can choose sensibly between them.

Cursor (hosted HTTP)

Cursor speaks HTTP natively. One URL, authenticated with your API key, nothing to install:

{
  "mcpServers": {
    "toolbox": {
      "url": "https://toolkitbackend-j69m.onrender.com/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer tb_live_<your_api_key>"
      }
    }
  }
}

1. Create a key on the API keys page.
2. Create .cursor/mcp.json in your project, or edit the global one in Settings → MCP.
3. Paste the block above, replacing the placeholder with your real key.
4. Reload Cursor. The endpoints appear in its tool list.

Claude Desktop (stdio bridge)

Claude Desktop speaks stdio, not HTTP. The bridge is a tiny local process that forwards tool calls to the same API. TOOLBOX_API_URL must be the API host (the value substituted below), not the website - the site does not proxy /api/v1.

{
  "mcpServers": {
    "toolbox": {
      "command": "npx",
      "args": ["-y", "toolbox-mcp"],
      "env": {
        "TOOLBOX_API_KEY": "tb_live_<your_api_key>",
        "TOOLBOX_API_URL": "https://toolkitbackend-j69m.onrender.com"
      }
    }
  }
}

Verifying it works

Ask the assistant something that needs a tool, such as "what are the MX records for example.com?". It should call dns-lookup and answer from the result. Each call appears in your usage log with the source marked MCP.

When something goes wrong