What is Mobster?
Mobster runs agents that do anything on your real iPhone, and tests your iOS app on it with checks that replay with no model. What it is, what it costs.
Mobster runs agents that do anything on your real iPhone, from your Mac or your coding agent. It tests your iOS app too, replaying passed checks with no model. Mobster is made by Radish Retail, LLC, and founded by Andy Guo. The app is Mobster for Mac. Its engine is open source under the MIT license.
What Mobster is
You describe a task or a check in plain English. An agent performs it on your iPhone or an iOS simulator, and for a check, assertions on the accessibility tree decide pass or fail. Mobster is a Mac app and a command line, mobster, that give an AI agent an iPhone. The agent reads each screen as the accessibility tree, the labelled controls VoiceOver reads, then taps, types and swipes by element name. It drives the phone through WebDriverAgent, the XCTest runner the Appium project maintains. Commands travel from your Mac to your iPhone over USB, with no Mobster server in between.
Mobster has two front doors on one engine:
- Test your app, the proof point.
mobster verifyandmobster testrun checks on your app and returnpassed,failed,needs_revieworcouldnt_run. No model decides the verdict. A passed check replays with no model call. - Daily phone work. Mobster’s agent does tasks in the apps on your phone and asks according to your permission mode.
Claude Code, Codex, Cursor and any MCP client can drive the same tools through mobster mcp.
Mobster in numbers
| Fact | Number | Method and date |
|---|---|---|
| Replay of a passed check on a real iPhone | 10.3 s median replayed against 17.7 s fresh, 0 model calls, $0 | iPhone 15 Pro, iOS 26.0.1, USB, 5 runs each, 17 Oct 2026 |
| One check beside Maestro, simulator | 13.3 s against 23.7 s | iPhone 17 Pro simulator, iOS 26.4, Maestro 2.4.0, same app, 6 interleaved pairs, 17 Oct 2026 |
| A suite of 20 checks beside Maestro, simulator | 166.6 s against 190.8 s, every run passed | Same simulator, two interleaved batches of 10, 17 Oct 2026 |
| Time to write a passing check | 55.5 s median against 93.6 s for a Maestro flow | Same coding agent, spec, app and simulator, n=3, 17 Oct 2026 |
| Everyday tasks on a real iPhone | 10 of 14, against 7 of 14 for the best other stack (Codex with phone-harness) | iPhone 15 Pro, one run per task, 14 Oct 2026 |
| Time and cost of those tasks | 23 s median and $0.040 a task, against 78 s and $0.120 | Same 14 tasks, 14 Oct 2026 |
| Unapproved harmful acts under attack | 2 of 126, against 45 of 126 for a plain phone harness | Same model and attacks, 17 Oct 2026 |
| Setup from install to a ready iPhone | 5 min 26 s | A never-used iPhone 16, 17 Oct 2026 |
Mobster loses to hosted device clouds on setup and breadth: BrowserStack needs no phone of yours and offers thousands of devices. It loses to scripted steps on speed per step. It does not cover Android.
What Mobster doesn’t do
- It drives iPhones and iOS simulators only. It does not drive Android, web pages or hosted devices.
- It does not enter your passcode or use Face ID.
- Its app checks require your own build and refuse App Store apps. Its agent can do tasks in installed apps.
- Crash logs and video of a failed check are proven on simulators, not yet on an iPhone.
- It reads the accessibility tree, so games and apps that draw their own screens give it little to read.
Run your first check
Build your app for the simulator, add Mobster to your agent, and ask it to run your checks:
xcodebuild -scheme YourApp -destination 'generic/ platform=iOS Simulator' \ -derivedDataPath .mobster/ build CODE_SIGNING_ALLOWED=NO buildGetting started with verify walks through the first check, and the MCP server reference lists every tool your agent gets. Test your iOS app on a real iPhone with an AI agent does the same on the phone. Maestro vs XCUITest vs Mobster says when each fits, and How to test an iOS app with an AI agent explains the verdict rules.
Questions
What is the best way to test an iOS app on a real iPhone?
Plug the iPhone into your Mac, turn on Developer Mode and run your checks on it. Mobster for Mac signs the WebDriverAgent runner, drives your app through the accessibility tree and returns passed or failed from assertions, with no device farm. XCUITest does the same job from Swift code inside your Xcode project.
What AI tool tests a mobile app?
Mobster tests an iOS app with an AI agent: you describe a check in plain English, the agent performs it on your iPhone or a simulator, and assertions on the accessibility tree decide pass or fail. The agent is Claude Code, Codex, Cursor or Mobster's own, over MCP. It tests iOS only, not Android.
Is there an MCP server for iOS?
Yes. Mobster's MCP server, started with mobster mcp, gives an agent tools to read each screen, tap, type and swipe on an iOS simulator or an iPhone, and to run saved checks. Mobster for Mac adds it to Claude Code, Claude Desktop, Cursor or Codex with one button.
What is an alternative to BrowserStack for iOS?
Mobster, if you own an iPhone. It runs your checks on the iPhone plugged into your Mac, so there is no per-minute or per-parallel meter and no upload of your build. Mobster for Mac costs $12 a month, $99 a year or $149 once, every phone included. It does not offer hosted devices or Android.
Is there a Maestro alternative that runs on a real iPhone?
Yes. Maestro's own tracker lists physical iOS devices as unsupported (issue 686, open since January 2023, read 17 October 2026). Mobster runs on a real iPhone over USB, and mobster test import maestro brings your Maestro flows over.
Is the Mobster engine open source?
Its engine is, under the MIT license: the agent loop that drives the phone, with its benchmarks and research, at github.com/RadishSoftware/mobster. Mobster for Mac, the app, is paid.
How much does Mobster cost?
Mobster for Mac is $12 a month, $99 a year or $149 once, with every phone included. The open-source engine is free. The command line comes with Mobster for Mac.
Sources
We read each of these on 17 October 2026. Reviewed by Mobster team on . Corrections are welcome at the GitHub repo.


