How it works
Your agent tests your app the way a person would, on the iPhone on your desk.
Install your build the way you already do: TestFlight, the App Store or Xcode. Zeptap doesn’t build or sign your app.
Describe the journey to Claude Code, Codex, Cursor or Zeptap’s built-in chat, in plain English.
The agent opens the app and taps, swipes and types through it like a user, while you watch on your Mac.
It tells you what broke and where. A coding agent can then fix the code and run the step again.
What your agent can test on a real iPhone
Your agent runs the journeys you’d otherwise check by hand, and reports what it saw at each step.
Test what you just built
Your coding agent changes a screen, checks it on your iPhone and fixes what doesn’t work. → Test your Claude Code iOS app on a real iPhone
Walk through onboarding
First launch, permission prompts, test-account sign-in. It flags confusing copy, dead buttons and errors.
Reproduce a bug report
It follows the customer’s steps, stops at the first screen that differs and tells you what it sees.
Smoke-test each build
Sign in, search, open an item, save it.
Check larger text and Dark Mode
It changes the setting, looks for clipped text, then puts the setting back.
Capture App Store screenshots
Every key screen of the real app. → Capture App Store screenshots with AI
Where AI agent testing fits among mobile app testing tools
Agent testing on your own iPhone covers the exploratory “does this actually work?” pass; scripted frameworks and device clouds still own large regression suites.
| Tool | Best for | Runs on | Needs |
|---|---|---|---|
| XCUITest, Appium, Maestro | Regression suites in CI | Simulators or a device cloud (XCUITest and Appium also on your devices) | Scripts you write and maintain |
| Device clouds | Many iPhone models and iOS versions | Remote phones | An uploaded build and a paid plan |
| Claude Code’s Simulator pane | Agent checks while you build | iOS Simulator only | Claude Code Desktop, Xcode, a paid Claude plan |
| Zeptap | Agent checks with real accounts and system screens | Your own iPhone, cabled to your Mac | The app on the phone, a Zeptap plan and your AI plan |
Anthropic’s docs say the Simulator pane “drives simulated devices only and can’t control a physical iPhone or iPad” (Anthropic, checked 14 October 2026).
What device clouds cost
Device clouds charge per user, parallel test, device or minute. Prices from each vendor’s pricing page, checked 14 October 2026:
| Service | Cheapest plan | Price | Billing |
|---|---|---|---|
| BrowserStack | App Live / App Automate | $39 / $199 a month | Annual |
| TestMu AI (formerly LambdaTest) | Live / Automation, 1 parallel | $39 / $199 a month | Annual |
| Sauce Labs | Live / Real Device Cloud | $39 / $199 a month | Annual ($49 / $249 monthly) |
| Maestro Cloud | Cloud | $250 per device a month | Monthly |
| AWS Device Farm | Metered / Unmetered | $0.17 per device minute / $250 per slot a month | Metered / monthly |
| Zeptap | One iPhone / up to five | $25 a month or $99 once (one iPhone); $199 once (up to five) | Monthly or one-time |
Clouds sell coverage of phones you don’t own; Zeptap uses the one you have, signed in.
What a real iPhone catches that a simulator can’t
A real iPhone runs the build your testers install. On it, your agent can also test:
- TestFlight and App Store builds, which the Simulator can’t install
- Real SMS sign-in codes
- The camera
- Links that hand off to Mail
- Real hardware and a real network
Apple notes that simulators “don’t replicate the performance or features of a physical device” (Apple, Running your app on simulated or physical devices, checked 14 October 2026).
Prompts to try
Paste one into Zeptap’s chat or your coding agent.
Chat · your iPhoneOpen our app as a new user, go through onboarding with the test account, and describe every screen that confused you.
Sign in, search for “running shoes”, open the first result and save it. Tell me if any step fails or feels slow.
Follow these steps from a bug report on the test account. Stop at the first screen that doesn’t match what the user expected and tell me what you see.
Show 3 more prompts
Set text size to the largest setting, open our sign-up and checkout screens, list anything clipped or overlapping, then restore the original size.
Install the latest TestFlight build, then check that sign-in, the notification prompt and the paywall’s Restore Purchases button all work.
Open the app on my iPhone, test every toggle on the settings screen you just changed, and fix anything that doesn’t stick after a relaunch.
You stay in control
You watch every tap live on your Mac and can stop the agent at any moment.
Nothing on the iPhone
No app, no jailbreak, no developer account for testing.
Your passcode never leaves your Mac
It’s kept on your Mac; the agent never sees it.
Your screen never touches Zeptap’s servers
Screenshots go only to the AI provider you chose. Use test accounts. Privacy
You set the stopping points
Tell your agent to check with you before purchases or messages to real contacts, and stop it from your Mac at any time.
your MacHonest limits
- It works from screenshots and on-screen text, not accessibility labels, so icon-only buttons are tapped by position.
- It runs one action at a time, when you run it. Zeptap has no built-in scheduler.
- The iPhone stays cabled to your Mac: good for checks while you build, not CI or a device matrix.
Start testing in three steps
Download Zeptap for your Mac (macOS 14 or later).
Plug in your iPhone and pair it. Zeptap walks you through it.
Ask your agent. Press ⌘L for Zeptap’s chat, or connect Claude Code, Cursor or any MCP agent in Settings › Connectors (connection guide).
Download for MacFirst hour free