ai-dashboad/flutter-skill

npm install -g flutter-skill   # or: brew install ai-dashboad/flutter-skill/flutter-skill | scoop install flutter-skill | docker pull ghcr.io/ai-dashboad/flutter-skill | dart pub global activate flutter_skill
Add MCP config: { mcpServers: { "flutter-skill": { "command": "flutter-skill", "args": ["server"] } } }
App SDK (Flutter): add flutter_skill dependency and call FlutterSkillBinding.ensureInitialized() in main() for kDebugMode

Usage

• •flutter-skill serve https://your-app.com # HTTP serve mode for CLI/automation/CI
• •flutter-skill nav / snap / screenshot / tap "Login" / type "email" / eval "document.title" / tools
• •Just talk to your AI: 'Test the checkout flow with an empty cart, then add 3 items and complete purchase'
• •CLI one-shots: flutter-skill explore <url> --depth=3 (autonomous bug finding), flutter-skill monkey <url> --actions=100, flutter-skill test --platforms web,electron,android

Key features

• ✓253 MCP tools (dynamic per page) covering explore, launch/connect, screen inspect, interaction, smart self-healing actions, assertions, visual regression, network mock/replay, API testing, multi-device parallel, a11y audits, i18n, performance, session, recording/export (Playwright/Cypress/XCUITest/Espresso/Detox/Maestro), auth, CDP browser control
• ✓Accessibility-tree snapshots (~200 tokens/page vs ~4,000 for screenshots; 87-99% fewer tokens) and batch explore_actions (5+ actions/call)
• ✓10-platform support with per-platform SDKs; claims 656/664 tests passing (98.8%) against a complex social-media app
• ✓Self-healing smart_tap/smart_enter_text/smart_assert with fuzzy match; monkey/fuzz testing; visual regression baselines; network mock/replay; API+UI testing; multi-device sync
• ✓Performance: tap 1-2ms end-to-end (vs 50-100ms for Playwright/Selenium) because it talks to the app runtime directly; 75/75 tools passed on heavy-DOM production sites with zero timeouts

Best for

AI-driven E2E testing of Flutter/mobile/web/desktop apps without writing any test code — natural-language QA, autonomous exploration, and monkey testing.

Caveats

• ⚠Native platforms (Flutter, React Native, Electron, Tauri, Android, iOS, KMP, MAUI) require injecting the corresponding SDK into the app; web CDP mode is zero-config but has fewer tests (141/156)
• ⚠Performance/token claims are the project's own benchmarks against a social-media demo app
• ⚠MCP server mode auto-launches Chrome; HTTP serve mode connects to an existing Chrome