Your agent takes the screenshots. You approve them.
Why this exists
Localized store screenshots are the part nobody finishes. Twenty languages means twenty runs through your own app, on a device set to a language you may not read, with a screenshot tool in the other hand. So most listings ship English screenshots everywhere, or the same five images from launch day. Your agent already has your repo open and already knows how to drive a build. This gives it the last mile: the actual captures, in the actual languages, in the exact sizes both stores demand.
How a run goes
- Your agent reads your app. It proposes which screens tell your story, and in which order. You correct it in chat, in your own words.
- One language first. The agent runs the flow once and shows you what it captured, so a wrong tap costs one run instead of twenty.
- Then every language you picked. It sets the app language per language, walks the same flow, and captures each screen.
- The set lands in Mokbi. It shows up under "From your agent" in the editor with a plan of which screenshot goes into which panel. You apply it in one click, and one Ctrl+Z takes it back out.
- Optionally, designed on arrival. If the agent sends a short brief about your app, the panel offers "Apply & design": the same AI Designer round the orange button runs, at the same price of 2 credits.
The capture runs on your machine
The MCP server has two halves and only one of them talks to us. Capture is local: it drives an Android emulator on your own machine over adb, and your source code, your build and the device never touch our servers. The remote half does one thing — stage the finished images in your project, behind a token you create in the editor and can revoke there. That token cannot publish to a store and cannot spend money.
Two checks that run on every set
A screenshot run that quietly captures the same language twenty times looks like a success and ruins a listing. So two things get verified:
- The language actually switched. After setting a language, the server reads back what the device reports. If that disagrees, the run stops and tells you.
- Two languages must not look identical. Identical screens mean your app isn't translated for that language. The threshold is measured, not guessed: two runs of the same language differ by 0.00%, British versus American English by 1.44%, Dutch by 5.10%, Japanese by 17.09%. Anything under 1% is flagged, with the status and navigation bars excluded because those change between runs and say nothing about translation.
Screenshots are the start, not the end
The same connection lets your agent draft your store text inside each store's character limits, translate the captions on your screenshots, and report what is still missing per language per store. You review all of it in Mokbi, where the store connection and the publishing checks live.
What it does not do
- It does not publish. Publishing happens in Mokbi, on your click, with the checks both stores need.
- It does not edit your design. An agent stages; you apply. That is why an open editor tab can never overwrite what your agent delivered.
- It does not test your app. If a flow hits a bug, you hear about it — fixing it is yours.
- iOS capture is not in there yet. That needs macOS and comes separately. Android capture is what ships today.