Human Path: Manual Installation

Use this path if you want full control over setup instead of delegating to an AI agent.

Prerequisites

• Flutter SDK installed
• A Flutter app you can run in debug mode
• Any MCP-capable assistant (Codex, Claude, Cursor, Cline, RooCode, etc.)

If you are deciding between terminal automation and assistant-driven flows, read CLI vs MCP first.

1. Clone And Build

git clone https://github.com/Arenukvern/mcp_flutter.git
cd mcp_flutter
make install

Binary path after build:

• mcp_server_dart/build/flutter_inspector_mcp

2. Add mcp_toolkit To Your App

From your Flutter app directory:

flutter pub add mcp_toolkit
flutter pub get

3. Initialize Toolkit In main.dart

import 'dart:async';
import 'package:flutter/material.dart';
import 'package:mcp_toolkit/mcp_toolkit.dart';

Future<void> main() async {
  runZonedGuarded(
    () async {
      WidgetsFlutterBinding.ensureInitialized();
      MCPToolkitBinding.instance
        ..initialize()
        ..initializeFlutterToolkit();
      runApp(const MyApp());
    },
    (error, stack) {
      MCPToolkitBinding.instance.handleZoneError(error, stack);
    },
  );
}

4. Run Flutter In Debug Mode

flutter run --debug --host-vmservice-port=8182 --dds-port=8181 --enable-vm-service

5. Configure Your AI Client

• Manual client setup
• Client-specific agent docs:
  • Codex
  • Cursor
  • Claude

6. Verify It Works

Try these prompts in your assistant:

• "List available tools from flutter-inspector"
• "Take a screenshot of the app"
• "Get app errors"

If multiple targets are found, retry with arguments.connection.targetId or prefer exact arguments.connection.uri from Flutter machine app.debugPort.wsUri.