Your agent writes the tests.
Your machine runs them.
Every night, for free.
Tapfleet turns your coding agent into a mobile QA engineer. It explores your app on a real device, writes the flows, and replays them nightly as deterministic steps — no model calls, no token bill, no device rental.
Three steps, then it runs itself
Connect your agent
Sign up, mint a project token, and point your agent at Tapfleet. It gets 28 tools: drive a device, read the screen, save a flow, read a failure.
claude mcp add --transport http tapfleet \
https://app.tapfleet.dev/mcp \
-H "Authorization: Bearer <token>"It writes the flow
Describe the feature in your own words. The agent opens your app on a simulator or a plugged-in phone, taps through it, and saves a flow that actually passed.
It replays for free
The flow compiles to deterministic steps. Nightly runs and pull-request checks execute those steps directly — the model is not in the loop, so the run costs nothing.
Everything after the test is written
Writing the first test is the easy part. Keeping hundreds of them honest, for years, is the work.
Zero-token replay
Regression runs make no model calls. A thousand nightly tests cost the same as one: nothing.
Your own devices
Runs on the machine your app is already built on. No device farm to rent, no build to upload.
Heals you approve
When a button moves, Tapfleet proposes the fix as a diff. Nothing turns green without a human.
Design diff
Attach a Figma frame to a step. Every later run compares pixels and shows what drifted.
Flake quarantine
A test that flips between pass and fail is pulled aside automatically, and released when it settles.
Pull-request gate
Every pull request gets a check with the runs, the screenshots and the reason it failed.
Your test suite gets cheaper as it grows
- —A model looks at the screen on every step, every night
- —Adding tests raises the monthly bill
- —Your build is uploaded to their cloud devices
- —Tests are healed silently, and pass anyway
- The model writes the test once, then steps out
- The thousandth test costs what the first one did
- Your build stays on your own machine
- Every heal is a diff you read and approve
The ones people actually ask
Keep using Maestro — Tapfleet is built on it. What you pay for starts after the flow file exists: nightly scheduling, run history, flake quarantine, the pull-request gate, design diffs, and a coverage map of the screens nobody tests.
Your agent is already on your machine. Give it a device.
Install the runner, connect the agent, and have a green nightly run before the end of the day.