AI-Powered UI Testing Platform

Stop Chasing
Broken Selectors

Replace brittle CSS selectors with stable semantic attributes. Tests that survive every refactor — forever. Powered by Claude AI + Playwright.

▶ View Pitch Deck ★ GitHub 📦 npm
<5KB
SDK Size
9
MCP Tools
3
Frameworks
3
LLM Providers
0
Cloud Deps
The Problem

CSS Selectors Break. Semantic Attributes Don't.

The moment a developer reorganises your DOM, every positional selector fails — not because the feature broke, but because a div moved.

❌ Fragile — CSS Selector
div.container > form
  > div:nth-child(2)
    > input.form-control

// Breaks on every refactor
✅ Stable — Semantic ID
[data-ai-id="login-email"]

// Stable forever ✓
Features

Everything You Need to Test Smarter

Claude MCP Playwright <5 KB SDK Zero Dependencies MIT License
🏷️
Semantic Attributes
Seven data-ai-* attributes describe what elements ARE — id, role, label, action, context, required, state. Survive every refactor.
📸
UI Snapshot API
getSnapshot() returns structured JSON with every element. Claude reads it to generate precise test scenarios automatically.
🔍
Auto-Tagging
Zero annotations? Auto-discovery finds inputs, buttons, and links on any page. Elements marked source:"auto" for AI confidence weighting.
⚡
Parallel Execution
Run up to 10 concurrent Playwright browsers with network mocking, session injection, and step-level retry on failures.
🖥️
Local Dashboard
Full dark-mode UI at localhost:3100. Snapshot any URL, browse element cards, view run history, export reports — no cloud needed.
🔌
Framework Adapters
Built-in adapters for React (useAiSnapshot hook), Vue 3 (composable + plugin), and Angular (injectable service).
MCP Tools

9 Tools. One Prompt.

Claude orchestrates the full pipeline automatically by chaining these tools. You never write test code.

01
get_ui_snapshot
Navigate to URL, inject SDK, return full element map
02
list_elements
Filter by role (input/action/nav) or source
03
generate_tests
AI generates TestScenario[] with optional hints
04
run_test
Execute single scenario with Playwright
05
run_tests_parallel
Up to 10 concurrent browsers + network mocking
06
get_results
Full step-by-step results with timing and errors
07
retry_failed
Re-run only failed steps with optional overrides
08
save_report
Export HTML · JSON · JUnit XML to any path
09
diff_snapshots
Detect UI regressions between deploys
End-to-End Workflow

One Sentence → Full Test Suite

"Test the login form at localhost:3000 and save an HTML report." — Claude handles the rest.

1
Snapshot
get_ui_snapshot scans all elements
2
Generate
Claude creates test scenarios via AI
3
Execute
Playwright runs scenarios in parallel
4
Results
Step-by-step pass/fail with timing
5
Report
HTML + JUnit XML written to disk

👻 See the Full Pitch Deck

14 slides covering architecture, SDK usage, MCP tools, dashboard, CI/CD, and live demos.

Open Slideshow →

Ready to test smarter?

Install the SDK, connect to Claude, and ask it to test your UI — in under 5 minutes.

npm install @phantomui/sdk
★ Star on GitHub 📦 View on npm ▶ Pitch Deck