mcp.example.json

{
  "_comment": "OPTIONAL native-MCP path. To enable, add \"mcpServers\": \"./mcp.example.json\" to plugin.json. The API key is injected by Claude Code from userConfig via ${user_config.api_token}. IMPORTANT: `flutterflow ai` reads FF_API_KEY (NOT FLUTTERFLOW_API_TOKEN — that var only feeds the legacy export-code path). Caveats: (1) the bare `command: \"flutterflow\"` is resolved via PATH, and this config hands that process the API key (FF_API_KEY) — so an executable named `flutterflow` earlier on PATH would run instead and receive the token. Prefer an absolute command path (e.g. ~/.pub-cache/bin/flutterflow), which also fixes the case where a GUI app can't see ~/.pub-cache/bin; (2) on macOS, ${ENV_VAR} substitution in plugin .mcp.json has been unreliable, but ${user_config.*} is resolved by Claude Code itself; (3) the server must run inside an INITIALIZED workspace — `--workspace \".\"` resolves to the launch cwd and fails with 'No .flutterflow/config.yaml found' if none exists. Create one with `flutterflow ai init` and point --workspace at it (an absolute path is safest).",
  "mcpServers": {
    "flutterflow": {
      "command": "flutterflow",
      "args": ["ai", "mcp", "--workspace", "."],
      "env": {
        "FF_API_KEY": "${user_config.api_token}"
      }
    }
  }
}