The Breaking Point
I had 500+ lines of E2E test code across three platforms. Every UI change broke dozens of selectors. Every framework migration meant rewriting test suites.
Then I tried something different: let AI control the app directly.
What is flutter-skill?
An MCP server with 253 tools that lets Claude, Cursor, or any MCP client:
- See your app through screenshots (31ms capture)
- Interact like a human — tap, scroll, type (1ms latency)
- Test across 10 platforms with natural language
- Auto-adapt when UI changes
Before vs After
Traditional (Playwright — 50+ lines)
await page.goto("https://app.example.com");
await page.waitForSelector("#register-button");
await page.click("#register-button");
await page.fill("[name=email]", "test@example.com");
// ... 50 more lines
AI-Driven (1 line)
flutter-skill test "Open registration, fill email and password, accept terms, submit, verify welcome page"
99.8% less code. Zero maintenance.
How it Compares
| Tool | MCP Tools | Platforms |
|---|---|---|
| flutter-skill | 253 | 10 (Flutter, RN, iOS, Android, Web, Electron, Tauri, KMP, .NET MAUI, CDP) |
| Playwright MCP | ~33 | 1 (Web only) |
| Appium | - | 2 (iOS + Android) |
Performance
- Tap latency: 1ms (near hardware limit)
- Screenshot: 31ms
- UI analysis: 2ms
- Compare: Selenium 100-500ms, Appium 200-1000ms
181 test scenarios across 8 platforms: 99% pass rate.
Get Started
npm i -g flutter-skill
flutter-skill init && flutter-skill demo
GitHub: github.com/ai-dashboad/flutter-skill
Open source, MIT licensed. Happy to answer questions!
Top comments (0)