Why I built it
Checking every page on a phone-sized screen is the QA step everyone agrees matters and nobody does consistently. I wanted mobile review to be as easy to run as a lint check.
What it does
- An MCP server that gives Claude Code 13 tools to drive iOS and Android device emulation
- Navigates pages, clicks, types, scrolls and captures screenshots
- Classifies every issue by category (layout, overflow, text, tap targets, navigation, accessibility, performance, visual) and severity
- Saves timestamped JSON and Markdown reports so you can compare runs
- Downloads Chrome for Testing on first run; Node.js is the only requirement
How it’s built
Pages are read as accessibility trees rather than raw screenshots, which describe a page’s structure in far fewer tokens and keep the agent’s context small.
The hard part
Making the output actionable. Early versions said things like “some text might be too small”. Forcing every finding into a fixed category and severity made the reports specific enough to work through.
Stack
TypeScriptMCPClaude CodeChrome for Testing