flutter-skill

flutter-skill acts as an MCP server that lets an AI agent control and test your running Flutter-powered apps without writing test code. It exposes an MCP-stdio interface so agents like Claude, Cursor, Windsurf, Copilot, and OpenClaw can send commands for navigation, interaction, and verification, while the server renders the app and reports results such as screenshots, accessibility trees, and state. You can also run in HTTP serve mode to drive automation via a REST API and CLI commands for actions like nav, screenshot, tap, type, and eval. The tooling supports 10 platforms through a unified interface, enabling a single MCP server to orchestrate across Flutter, React Native, Electron, Tauri, Android, iOS, and more. With zero-config and natural-language prompting, you can describe goals in plain English and have the agent perform the steps end-to-end with minimal setup.

Prerequisites:

• Node.js and npm (to install the Flutter Skill CLI via npm) or an alternative package manager as described below.
• Basic knowledge of MCP configuration (mcp.json) and how to run MCP-enabled tools.

Installation options:

Option A: npm (recommended for quick start)

# Install globally
npm install -g flutter-skill

# Start the server (example)
flutter-skill server

Option B: Homebrew (macOS)

brew install ai-dashboad/flutter-skill/flutter-skill

Option C: Snippets for other methods

# Docker
docker pull ghcr.io/ai-dashboad/flutter-skill

# Dart pub (if available via pub global activate)
dart pub global activate flutter_skill

Option D: Quick MCP config snippet (used by agents)

{
  "mcpServers": {
    "flutter-skill": {
      "command": "flutter-skill",
      "args": ["server"]
    }
  }
}

Note: The README mentions multiple install methods (Homebrew, Scoop, Docker, IDE integrations, etc.). Choose the method that best fits your environment and deployment model.

Tips and considerations:

• The MCP server supports both the stdio MCP mode and an HTTP serve mode for remote control. Use the server command with the appropriate flags if you need port/host customization.
• Typical CLI commands you’ll use once the server is running include: nav <url>, screenshot [path], tap <selector>, type <text>, eval <js>, tools, and call <tool>.
• When integrating with agents, you can paste the MCP config snippet into your agent’s config (e.g., claude_desktop_config.json, Windsurf config, etc.).
• Environment variables like FS_PORT and FS_HOST can override defaults for the HTTP interface.
• If you encounter issues, verify your app is running and accessible, ensure the Flutter Skill server has network access, and check that the MCP config references the correct command and arguments (as shown in the mcp_config example).
• For multi-platform testing, you can rely on the 10-platform support to drive actions across Flutter, React Native, Electron, Tauri, Android, iOS, Web, and more without changing your test scripts.