Agentic UI Testing.
Open source, human-readable code, easy CI integrations, and powerful CLI and desktop app.maestro.dev San FranciscoJoined October 2020
It's so much better when your own coding agent opens the app and checks that what it built works. That's what Maestro MCP is for: eyes and hands for @claudeai, @cursor_ai or Codex.
Your coding agent has a better work-life balance than you.
Journal: "finished work early, got a workout in, read a few chapters..."
Give your coding agent eyes and hands to control your app. Maestro MCP.
This is what automated UI testing looks like in 2026. An agent taps through every settings screen in Sleep Cycle, handles a permission dialog and a failed download, and Maestro records it all as a rerunnable test.
@claudeai code caught its own flaky test step, added a wait for the results to settle, and stabilized the test itself. All driven through the Maestro MCP.
Automate your UI tests any way you want: with your coding agent, tap by tap in Maestro Studio's visual IDE, or in your own editor and run from the CLI.
All 3 in Maestro. OPEN SOURCE.
Now you can launch a specific Android system image with --device-os in Maestro CLI 2.10.0.
The flag takes a full system image path, so your run boots the exact image you need, including images with the Google Play Store libraries.
Release notes: maestro.dev/blog/maestro-c…
In loving memory of driver.findElement(By.xpath("(//android.widget.EditText)[1]")).sendKeys("Review PRs");
Same step today: inputText: "Review PRs". Written by your agent through Maestro MCP, from one sentence.
Hot take: the best UI test is the one you never had to write.
Go through the flow yourself, record it, and you get something simple, deterministic, and much closer to how users actually interact with your product.
you just finished the feature. instead of tapping through 6 screens to get to it, ask the agent to go check it. the flow it just ran is your regression test.
"Test the Monefy app, add a $50 expense for Food, $20 for Transport, $30 for Entertainment, then check the spending chart."
That sentence was the entire test plan. Claude Code opened the app, closed the paywall that popped up, entered all 3, and checked the chart read 50 / 20 / 30.
Grok testing Grok.
One prompt: QA the Imagine feature, generate an image, confirm it can be turned into a video.
Grok Build opened the Android app through Maestro MCP, typed the prompt, waited for the 4 stills, opened one, tapped Video and checked the conversion started. Pass, end to end.
25K Followers 3K Following🤹♂️ (Former) Director of Eng at @Nubank, @Apple, @AmericanExpress. 🇧🇷 in🗽.
I share career growth strategies, leadership insights & my many reads. 📚🐛.👇
24K Followers 347 FollowingA software R&D agency. We work with founders in various fields to build custom products or developer tooling. Available for hire.
25K Followers 1K FollowingBuilding an indie app business in public: @habitkit @focuskitapp
Sharing my learnings about development, marketing and sales.
1 Followers 6 Following🚗 AutoBazar — a modern car marketplace in Croatia. Making buying and selling cars easier with smarter search and better listings.
13 Followers 82 FollowingHelping companies transform ESG & non-financial data into clear, compliant, and actionable insights — turning reporting into real impact. 🌍✨
91 Followers 933 FollowingFounder of Spicy Brain, a neurodivergent friendly reminder platform for running your life.
Devout lover of Monster Hunter and anime
ADHD Haver