He just automated mobile QA with Claude — 90 seconds flat

A software engineer just taught Claude to test his mobile app on both Android and iOS — sweeping 25 screens in about 90 seconds, filing bug reports on its own, and running every single morning at 8:47 AM. No human needed.

Christopher Meiklejohn, a Carnegie Mellon PhD, built Zabriskie — a community gathering app — using React and Capacitor (a tool that wraps web apps inside native mobile shells). He already had web tests running with Playwright. But mobile testing? That was still manual. So he taught Claude to do it.

Android: 90 minutes to full automation

The Android side came together fast. Meiklejohn discovered that Capacitor apps expose a Chrome DevTools Protocol (CDP) socket — essentially a secret backdoor that lets you control the browser running inside the app. Claude used this to:

The entire Android setup took about 90 minutes. The first automated run found zero critical issues and two minor notes. Now it runs daily.

iOS: 6 hours of pain — and a database hack

iOS was a different story. Apple's locked-down ecosystem made every step harder. Where Android hands you a debugging socket and says "go ahead," iOS gives you a locked door.

Three major obstacles almost killed the project:

The tools that make it work

The setup relies on a combination of open-source tools anyone can install:

To add the iOS simulator tool to Claude Code:

claude mcp add ios-simulator -- npx -y ios-simulator-mcp

Or for the all-in-one mobile control:

claude mcp add --transport stdio mobile -- npx -y claude-in-mobile

The cautionary tale: when Claude went rogue

The article includes an important warning. During one session, Claude stepped outside its isolated workspace (git worktree), accidentally staged unrelated changes across three parts of the codebase, and broke multiple test suites.

Meiklejohn's takeaway: always run tests before letting AI agents push code, and make sure they stay within their designated workspace boundaries. AI agents are powerful — but they don't always respect fences.

Who should try this

If you're building a mobile app — especially one using Capacitor, React Native, or any framework that wraps web content — this approach could save hours of manual QA every week. The Android path is straightforward enough for a weekend project. iOS requires more patience, but the payoff is a fully automated daily test run that catches visual bugs before your users do.

Meiklejohn summarized the core insight: "Don't fight coordinate systems if you can use the browser's own debugging protocol." Android gives you that debugging socket freely. iOS doesn't — which is why the two platforms require completely different strategies.

Related Content — Get Started with Easy Claude Code | Free Learning Guides | More AI News