Maestro testing and AI agents: script the flows you know, let an agent find the rest.
Maestro facts from its docs and pricing, checked October 12, 2026.
Flows and agents side by side
| Maestro | AI agent on Phonebox | |
|---|---|---|
| You write | YAML flows, step by step | A plain-language task |
| Best at | Known journeys, fast regression checks | New screens, edge cases, bug reproduction |
| Runs on | Local emulators and devices, or Maestro Cloud ($250 per device a month) | Cloud Android phones, $0.06/min, billed per second |
| Reports | Pass or fail per step | What it saw, quoting the screen text |
A flow, and the same check as a task
A Maestro flow for a checkout journey:
appId: com.example.shop
---
- launchApp
- tapOn: "Sign in"
- tapOn: "Email"
- inputText: "qa+checkout@example.com"
- tapOn: "Continue"
- tapOn: "Basket"
- tapOn: "Checkout"
- assertVisible: "Order placed"The flow checks exactly these steps, in this order. Ask an agent the same thing in a sentence, "sign in with the test account, check out the basket and confirm the order is placed", and it can also notice what the flow never checks, such as a promo code that didn't apply, because it reads every screen on the way.
Turn what the agent finds into flows
This prompt closes the loop: the agent explores, and each bug it confirms comes back as a flow, written with the screen text it actually saw:
Explore the checkout flow of the app on my Phonebox phone and report
anything broken, quoting the screen text. For every bug you confirm,
write the exact steps you took as a Maestro flow in .maestro/, using
tapOn, inputText and assertVisible with the text you saw on screen,
so the bug stays fixed. Park the phone when you finish.Run the new flows in CI with Maestro on an emulator or Maestro Cloud. Phonebox phones don't expose ADB, so Maestro itself doesn't run on them.
What each costs
| Use | Cost |
|---|---|
| Maestro flows on your own emulator | Free (open source) |
| Maestro Cloud, one device | $250 a month |
| 20 agent explorations of 10 minutes on Phonebox | $12 of phone time, plus your model's tokens |
Questions
Can AI replace Maestro tests?
Not for regression checks you run on every commit: scripted flows are faster, cheaper and deterministic. An AI agent is better at exploratory checks of new screens and reproducing bug reports, and it can write each bug it confirms as a new Maestro flow.
Does Maestro have an MCP server?
Yes. The Maestro CLI includes an MCP server (maestro mcp) that runs locally and can send runs to Maestro Cloud.
Can an AI agent write Maestro flows?
Yes. A coding agent that explored the app on a Phonebox phone can write the steps it took as a Maestro flow using tapOn, inputText and assertVisible with the exact screen text it saw, ready to run in CI.