Vee3

Connect Vee3 MCP to Pi

Published: July 8, 2026

Ask your agent

Recommended

Copy and paste this into Pi chat and let the agent walk you through setup.

Agent Instruction

Connect to Pi

Pi needs the pi-mcp-adapter extension before MCP config works.

  1. Install the adapter and restart Pi:

    pi install

    Requires Pi coding agent. Restart Pi after this finishes.

    pi install npm:pi-mcp-adapter
  2. Add MCP config - pick one location:
    • .mcp.json in your project root (shared across hosts)
    • ~/.config/mcp/mcp.json for all projects
    • ~/.pi/agent/mcp.json for Pi-only global config
    • .pi/mcp.json for Pi-only project override
  3. Paste the config from the copy block below under mcpServers.
  4. Set VEE3_API_KEY in your shell before starting Pi when using ${VEE3_API_KEY} in headers.
  5. Verify with /mcp or /mcp reconnect vee3.
  6. Test - e.g. “Use mcp to search for tools, then capture a screenshot of https://example.com”.

mcp.json (HTTP - recommended)

Recommended

Uses ${VEE3_API_KEY} - also supported as $env:VEE3_API_KEY in headers.

{
  "mcpServers": {
    "vee3": {
      "url": "https://mcp.vee3.io/mcp",
      "headers": {
        "VEE3_API_KEY": "${VEE3_API_KEY}"
      },
      "lifecycle": "lazy"
    }
  }
}
Direct Pi tools (optional)Show

You can register tools as first-class Pi tools. This uses more context than the proxy.

mcp.json (directTools)

Replace the vee3 entry with this variant, or merge directTools into your HTTP entry.

{
  "mcpServers": {
    "vee3": {
      "url": "https://mcp.vee3.io/mcp",
      "headers": {
        "VEE3_API_KEY": "${VEE3_API_KEY}"
      },
      "lifecycle": "lazy",
      "directTools": [
        "{tool_name}"
      ]
    }
  }
}
Alternative: stdio via mcp-remoteShow

Use only if direct HTTP does not work. Requires Node.js and npx.

mcp.json (stdio bridge)

Replace the vee3 HTTP entry with this stdio entry.

{
  "mcpServers": {
    "vee3": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote@latest",
        "https://mcp.vee3.io/mcp",
        "--header",
        "VEE3_API_KEY: ${VEE3_API_KEY}"
      ],
      "lifecycle": "lazy"
    }
  }
}

Proxy tool by default

Pi exposes one mcp tool instead of loading every Vee3 tool definition. Ask Pi to search via mcp or enable directTools for the screenshot tool.

Pi adapter docs: pi-mcp-adapter. Do not commit literal API keys.

Verify the connection

Run /mcp or /mcp reconnect vee3 and confirm vee3 appears. Then ask the agent to call meta-tools.list_groups or search the catalog with meta-tools.search via the mcp proxy tool.

Customize which tools appear

From the dashboard, open Customize MCP to turn capability groups on or off. Changes apply to your MCP server on the next connection.

Next steps