Four things we added in the last weeks. They all work today, and they are all beta: expect rough edges, and tell us when something breaks.
Connect your own app on your own device to your Claude Code session. Claude sees the screen, the texts on it and the notes you send from the app — and can tap, type and scroll through the app for you. You switch it on in your dev build and tap "Connect to Claude" in the app.
Limits: Only your own device and app. Destructive actions (delete account, pay, sign out) are blocked on the device. Available in all four SDKs from 1.2.0.
Claude can draft a test order for real testers over MCP: which app, which platforms, how many testers, which tasks, which reward. Drafting is free and stays private.
Limits: Before anything is published or costs money, Claude shows you the price preview and needs your explicit yes to exactly that total. If the order changes, the yes is void.
Write down what testers should check — a few concrete steps per task. Testers see the checklist in the app under "What to test", tick each step off with ✓ or ✗ and add a note.
Limits: You see per task how the latest human and AI runs went, step by step.
When you mark a bug as fixed, an AI tester can walk through the bug’s acceptance steps again — on the fixed build only, not on whatever happens to be running. Passed: the bug counts as verified. Failed: the bug reopens with the step that failed.
Limits: Flutter and Expo only, and only with the AI tester enabled. For high and critical bugs a human still confirms.
Create a project account, integrate an SDK from version 1.2.0 and set up the TesterPayKit connection in Claude Code. Paid orders are not available yet.