I Made Claude Code My iOS App Tester - It Found a Real Bug

James Pelton
31 followers
10 Video Views·Jul 6, 2026
Manual app testing eats hours. Every login flow, every form, every screen tapped by hand. In this video I put Callstack's open-source agent-device to work driving a real React Native app and handling the repetitive QA flows for me.
Thanks to Callstack for sponsoring this video.
🔧 Try agent-device (free, open source):
https://github.com/callstack/agent-device
👋 Follow on X: https://twitter.com/jamespelton18
👥 Community: https://www.skool.com/6-million-dollar-app-2514/about
🛠️ Want my team to build your app? https://passive-apps.com
🙏 Proverbs 14:23 — "In all toil there is profit, but mere talk tends only to poverty."
⚠️ Transparency: This video is sponsored by Callstack. agent-device is free and open source. I only feature tools I'd actually use.
So I have been building a lot of apps
as part of my app program
and AI has made the building the apps pretty easy.
But what it hasn't helped too much with
is the testing of the apps.
Whenever I wanna test one of these apps,
I have to do the same login flow by hand.
I have to launch a simulator, type the email,
type the password, sign in,
make sure whatever feature I'm working on actually works.
And I have to do this over and over.
It's very time consuming.
So today I wanna look at some software
that Callstack put out called Agent Device
where my Cloud Code can connect to it
and it can actually control iOS apps.
So I should be able to test my iOS apps in Cloud Code now.
So this is Callstack
and they open sourced a tool called Agent Device,
but it lets AI agents control real iOS,
Android, TV, and even desktop apps.
But it doesn't do it by guessing at pixels,
but it actually reads the actual accessibility tree.
So the same thing that a screen reader uses,
and then it taps elements by reference.
It's MIT licensed, completely free,
and it works with any agent that can call CLI tools.
So you can use it with Cloud Code, Cursor, Codex,
whatever you wanna use.
You might notice this is a different setup for me.
My apps, I build on my laptop.
So I'm recording from here so you can kinda see what I do.
So first thing, I wanna get the MCP server set up.
So MCP servers are basically how like Cloud Code,
AI can connect to things.
So I'm gonna add one for Agent Device here.
And then in Cloud Code, when I do slash MCP,
you can see Agent Device is now an MCP server on here.
So now Cloud has tools to open my app,
read the screen, tap, and type on a real simulator.
So that's the whole setup.
So I'm telling Cloud just to open BuildSync
in the iOS simulator, test the login screen,
go to sign in, fill in the email,
test password so it's gonna probably fail
saying invalid password, and let's see what happens.
All right, so put in the email, the password,
and says login failed, and Cloud actually had a suggestion.
It said that's not a very good message, like login failed.
It should say invalid username or password.
So that's one thing I could fix.
Honestly, I could probably just get Cloud Code going
and like, hey, just go test out everything in the app
and fix what you see, but this is the first thing.
So actually, let's go ahead and fix this first.
So we'll have Cloud fix its first problem that it found
and then test it out, make sure it's fixed now.
All right, so it took Cloud about two minutes
to go make the code changes,
and now it does an invalid username and password
and gives a better error.
So already, this is one bug that we were able to fix
using this new agent device.
Okay, I wanna give it another bug to fix.
I noticed that our tabs are messed up here at the bottom,
so let's go ahead.
I'm gonna just take a screenshot and say,
let's fix the tabs.
Let's fix the tabs and then test to make sure it is working.
Cloud sees the problem, it's got these placeholder icons,
and it will get going fixing it and then testing it,
make sure it's good to go.
All right, and it fixed that as well.
This is gonna really help my workflow.
Those of you who are in the app program,
this will help a lot with QA,
because what happened before
was I could tell Cloud to fix something,
but then it couldn't verify it was fixed,
so it would say I fixed it,
then I would have to go look at it and say,
no, it's not actually fixed,
and I'd have to paste back a screenshot or something.
Really help out my workflow quite a bit.
If you are doing any kind of app building,
I think you need to go check this out.
I'll leave a link down below.
But connect the MCP server to your Cloud Code
and get going on it.
It's gonna be great for testing for me.
And shout out to Callstack.
This is helping me tremendously,
so thank you guys for building this.
Thank you for making it open source.
If you guys are wanting to get more into app building
or using AI to build an app,
go check out the AI Income Builders Club.
So I have a full walkthrough about building apps using AI.
I'm definitely gonna add this as a module in there,
because it's gonna be super helpful.
So thank you guys so much for watching.
I'll be back at my big computer tomorrow.
God bless you guys, and have a great rest of your day.
