We built a tool that lets Claude Code autonomously test your entire mobile app. It navigates your app, opens screens, reads the debug logs, finds performance problems, and hands you a structured report. All it takes is one prompt.

Free & open source. Star the repo ⭐👇

This looks really useful for React Native apps. Might try this instead of my usual mobile testing setup.

Give it a try: npx @swmansion/argent init

does it work with codex too

Argent works with any coding harness that supports MCP servers including Codex, Cursor, Claude Code, OpenCode etc.

I have a question: Can it handle edge cases, like rapid tapping, clicking on wrong button, or doing something wrong intentionally? Because testers test apps in a way that user experience remain good.

The missing context is that this works great for deterministic flows but production mobile apps with dynamic UI states or flaky networks still require human verification loops to avoid false positives in the report.

Argent won't replace a human but can drastically help during local development of mobile apps.

loove argent. Would be brilliant to have a best practices case skill fine tuned for UI automation testing.

Argent can do it on its own! It can group repetitive UI sequences into "flows" and reuse them in future sessions.

does it work for expo apps? Like will it test the ios and android versions in one go?

It works with Expo apps. Argent will test iOS and Android one after the other only if you prompt it to!

This is 🔥 Autonomous mobile testing with one prompt is exactly the kind of AI agent magic we need. Huge time-saver for indie devs and teams. AI that lets educators generate full mock tests, study materials & smart analysis in seconds instead of hours. Love tools that turn one prompt into real work. What’s the biggest pain point you’re solving for mobile teams? Let’s connect!

@brandtnewlabs Nice work! When to use agent-device and when argent? Or are they more or less the same?

Cool! Is it faster and different than maestro?

Argent is a tool you'd use during local development to give coding agents superpowers including profiling & debugging mobile apps. Maestro is an e2e testing framework where you define a flow using yaml and then run a deterministic test runner to check your assumptions.

Thanks!

Does it catch native issue or only react native related issues?

Not only React Native. Argent comes with tools for native development like native debugger or native network logs.

@kacperkapusciak I am curious if I can create maestro flow to cover my app e2e! It would be amazing to use this to explore and create tests! Well done, once again 👏

@kacperkapusciak You can prompt your favourite coding harness to explore your app using Argent and let it write Maestro tests based on those results.

i'd care less about finding the bug once and more about rerunning the exact weird path after i fix it.

11 screens. 3 crash logs. still didn't find the budget chart bug.

@kacperkapusciak How does this compare to Agent Device 👀

Very cool! How does this differ from Maestro in practice? Also, do you have plans on supporting real devices connected by USB?

this is the missing piece. e2e testing on mobile has been the part where claude code falls off vs web. what's powering the simulator under the hood, xcode + adb or something custom?

Using debug logs in the loop is the interesting bit here. When the app hits OS permission sheets or login state changes, does the report separate app bugs from test-environment noise?

Argent brings tools and feeds the output into coding harnesses. Frontier models like Opus 4.7 or GPT 5.5 are incredibly capable of distinguishing between system alerts and the application they are meant to test.

That makes sense if the harness keeps the alert context explicit. The edge case I’d still watch is repeated permission sheets after login/session reset, because those can look like app state unless the report preserves the OS prompt boundary.

will be very useful, but i think Maestro are catching up fast 🫢

I tried it and it’s amazing work you guys did! But I couldn’t help but notice that it was extremely slow between argent tool calls <-> agent back and forth, every time waiting for Claude to do 1 small next step. Any tips on this?

qa engineer here — does it handle auth flows or mostly happy path? that's where my current setup dies

I built this at a hackathon this weekend and won. But I think this capability should be extended to real devices as well, since the iOS simulator is marginally less capable. iOS simulator is a "simulator" not an emulator.

you can pay 2 employees in qa for the token cost

How much token does it consume for testing roughly 6-7 screens with flatlist .

simulator only?

Argent can interact with iOS simulator and Android emulator.

I've seen many tools like this recently

does it work for flutter ?

Hi there tried your message onboarding ui I am new did I try