Here's Claude tapping, scrolling, finding bugs and fixing them. Testing the entire app experience on iOS. No Xcode. No manual steps. FlowDeck gives your agent eyes on the simulator. Post + Video:

$60 a year for a UITest wrapper? No thanks.

Far from a UITest wrapper, but okay. Your call. :)

@itsfallah got claude to remake this in like 5 mins lol

@afterxleep I thought about doing this too. Gotta appreciate the grift.

Sees the bugs, fixes them, now ship it. handles the App Store submission.

How’s this different from writing one-time "deterministic" playwright test cases? In this case, you will be burning tokens every time you run You can't deterministically confirm if all the test cases are passing or not I see the appeal but not long-term

Playwright takes screenshots and taps by coordinates, while FlowDeck uses your app's accesibility IDs by default. That means theres a HUGE token reduction when not processing images. Still, if you dont have IDs, it can use screenshots.

Let me play devil’s advocate to understand the benefits (no ill intent). Fundamentally, I’m assuming the codebase is written with proper IDs and classes, and is well structured (whether AI- or hand-written). 1/

Once we have a decent codebase, writing test cases will be a one-time cost. Can use AI for that, but going forward, we only need to run those tests. As we incrementally add features, add corresponding test cases. think this leads to a much larger reduction in token usage overall.

This doesn't replace UITests. Write those. What it does is find what UITests can't. Edge cases, layout bugs, rendering issues your tests never covered. The agent actually sees the screen. Across devices, across platforms. It's the QA pass you never have time to do.

Sorry if UTs distracted the conversation. I'm saying - we write playwright using Claude and during that time we also take screenshots of what each screen in all possible combinations can look like. These are all test cases - screenshots+clicks. These are repeatable caes. 1/

To that extent, we can test various screens and dimensions. It does a pixel comparison, so once it's rendered, we can take screenshots. Can even see rendering time. I'm currently try to evaluate for what case we genuinely need llms. Not trying to discourage your work! It's good.

Yea, sure. You can always put something custom together. Thats perfectly valid

@PaulSolt Xcode is still needed to install and build the app

@PaulSolt Yes

I just prayed for this

you can stop now :🙏

I just finished a meeting and was talking about this, especially as we had a bug that was due to older api on user who had older phone

This sold me, just bought myself a 1 year license.

🖤 thank you for the support!. What are you building?

I’m a staff iOS eng @IntelyCare and would like to introduce this to the team.

@IntelyCare Awesome! Ping me if you need more info. Here's some demo videos to share:

@IntelyCare Will do, and thanks for personally reaching out.

I built a skill for it to use simulator how is this different?

FlowDeck is a fully native CLI that’s faster and more reliable than letting the agent use xcodebuild directly.

That’s pretty sweet

Prompt to Claude: "Recreate flowdeck". Wait for an hour. Profit!

Claude doing full iOS QA with no Xcode? Bruh that's wild. QA teams sweating and lowkey jealous lol

We’re a lot closer to witness AI building, testing and deploying secure apps

How is this different from just the mobile-mcp you can install and give Claude?

Agent finds the bug, fixes the code, then taps through the app to verify. The QA loop just collapsed from 'file a ticket and wait' to 'fix and prove it' in one session.

Love the approach, definitely another step in direction. The major issue for me is fixing issue in such way it burns a LOT of tokens and a time I tried to QA the build through the MCP. - Time consuming, like x3-x5 comparing to regular QA with specific behaviour expectations. - Tokens and sometimes it stucks in loops because of the different reasons

Daniel, can it test extensions like the keyboard?

It should work but haven’t tested them yet. Grab a trial and try it out!

Genuinely asking, what's the difference between using this or the Blitz MCP or ios-simulator-mcp?

Giving Claude eyes on the simulator is genuinely a game changer for solo devs. No more manual tap-through testing.

IOS App: Grindmate: Personal Assistant. 100 Tester only Free Full Functions no paywalls.

How is it different from claude’s own computer use mcp?

Claude will capture your full screen shots, costing you time and tokens. This is designed for Apple Development and uses Apple tools and accesibility IDs for minimum token usage. It's also a CLI, so it works everywhere, no servers needed.

@PaulSolt Looks really interesting! Wondering what’s the extra this adds on top of something like XcodeBuildMCP ? (Seriously curious)

@PaulSolt You get FlowDeck interactive mode (Terminal UI), macOS automation, 100x faster in test discovery, and xcode project parsing. Also its 3x more efficient in Token Usage compared to it. Here's a good read:

@PaulSolt is the discovery path also - screenshot, find a tap coordinate, tap, screenshot, over and over? seems that's what other tools are doing and its slow... seemingly.

@PaulSolt Screenshots are a fallback. By default, it exposes the accessibility tree to the agent in real time, and falls back to screenshots only when there are issues.

flowdeck for ios testing, no xcode, end-to-end automation wins

impressive👍

Holy token hog

What does it do better than @xcodebuildmcp ?

Flowdeck is a game changer. Codex works with it trough the Flowdeck skills to run tests and sim. It also saved hours to quickly download an old sim I needed to find a bug!

The real unlock here is closing the edit-verify loop without a human in the seat — agent changes code, then visually confirms the fix on a real simulator. That's the gap that made mobile dev resist automation way longer than web.

Thats great, Im building something similar actually but using AI for generating the test steps only and my engine responsible for executing the tests but it supports only Flutter for now

That's insane, HF

test suites only cover what you imagined could break agents exploring UIs without a script will find the stuff nobody wrote a test case for most production bugs live in that gap

Bro, I need this but for a regular web app. Playwright is not enough.

I know.

Watching Claude tap through an iOS app to find bugs is kind of wild. FlowDeck giving the agent a simulator view is a clever approach

Stunning! Is it on GitHub? Can it generally be used even for non Xcode related coding?

Its only available for native development.

What bugs is it fixing in the weather app

It found an issue with city sorting while testing. Here’s the full video

Another weather app … cool

oh yes, we need more...