Expo Go for agents: a phone your AI can actually use
Expo Go is a lovely loop — for a human. Scan a QR code, and your code runs on the phone in your hand. Your AI agent has no hand, no phone, and no way to scan anything. PreviewAPK gives the agent the same loop on a phone of its own: a cloud Android it can push builds to, drive, watch, and debug — then hand to you.
Try it with a live device
PreviewAPK is in private beta. Get early access to test your own builds.
Why agents can't use Expo Go
It lives on a physical phone
Expo Go is an app installed from the Play Store. The whole loop — install, scan, hold, watch — assumes a person with a device on the same network as the dev server. Nothing in it answers to an API, a CLI, or an MCP tool call.
It only runs part of your app
Expo Go ships with a fixed set of native libraries. The moment your project adds a native module that isn't in the box — payments, maps SDKs, Firebase, your own native code — Expo Go errors out, and Expo's own answer is to switch to a development build. Agents hit this wall constantly, because real apps have native code.
There's nothing to hand over
When the agent finishes, there's no session you can open, watch, or take over. You get a claim — "it works" — and the job of testing it yourself.
What the agent gets instead
A device slot on PreviewAPK is a cloud phone, priced flat at $19/month per device. The agent pushes any APK — an EAS build, a development build, a release candidate — with one CLI command or an MCP tool call, and the phone boots it in seconds. From there it has the full set of senses and hands:
- Drive: tap (including by a button's label), swipe, hardware keys, and real text typing.
- See: screenshots it can actually look at, plus the live screen streamed to a browser.
- Understand: the screen's UI hierarchy — elements with labels and positions, not pixel-guessing.
- Hear: the app's debug log, including the crash trace with its reason attached.
It tests and debugs its own work in a loop — and when it says "done," there's a session you can open to check. That's the agent testing loop in full, and the MCP reference lists every tool.
Then it hands you the phone
The same phone is yours. Open your portal and you're watching the same session your agent just verified — tap through the feature yourself, no Android Studio, no SDK, no install. The agent and the human take turns driving one device; it never belongs to just one of you. And if a client needs to see it, the session's view-only link is a byproduct — send it, they watch, they never drive.
Not only for Expo projects
Expo apps arrive as APKs, and so does everything else: bare React Native, Flutter, Kotlin, a client's white-label build. If it builds to an APK, the agent's phone runs it — which makes PreviewAPK a fine Expo Go alternative for humans too, and the developer & agent guide covers the CLI, API, and MCP setup end to end.
FAQ
Can an AI agent use Expo Go?
Not in any practical way. Expo Go is an app that runs on a physical phone: a human installs it, scans a QR code, and holds the device while it runs. An AI agent has no phone, no camera for the QR code, and no way to tap, screenshot, or read logs from inside it. Agents need a device that answers to an API — which is what PreviewAPK provides.
What does an agent need to test an Expo app?
A built APK — from EAS Build or a local build, including development builds — and a device it can drive. Push the APK to PreviewAPK and the agent gets a cloud phone: it starts the app, taps and types, reads screenshots and the UI hierarchy, and reads the app's logs when something breaks.
Does this work with Expo development builds?
Yes. A development build is an APK like any other, and it runs in full — including the custom native modules that Expo Go can't load. Many teams keep Expo Go for the first prototype, then have their agent test development builds on PreviewAPK.
Is PreviewAPK affiliated with Expo?
No. PreviewAPK is an independent product. Expo and Expo Go are trademarks of their respective owners; this page compares workflows, nothing more.