Git Qa
Open-source QA runner where an AI operates your test sheet and a person signs off each
- Pricing
- Open source
- Platforms
- macOS, Windows, Android
- Version
- 0.2.0
- Listing updated
- today
Similar MCP servers
About Git Qa
Git Qa is an open-source MCP server for developer tools. It is available on macOS, Windows, Android, iOS, Web, CLI, API and self-hosted.
The README describes git-qa as a runner that works through a TSV test sheet while a human watches the live screen and presses D to pass or F to fail each case. Verdicts are split so that a case the person signed off is recorded as VERIFIED with their handle, while cases the AI passed alone stay AUTO_PASS, which keeps AI-only passes distinct from human-verified ones.
The AI side can drive Android devices and emulators over adb, web pages in several browsers, macOS and Windows desktop apps, and iPhone or iPad through WebDriverAgent. Each finished case writes a run.json record and a screenshot, with an optional video. Test sheets are plain TSV files that can be edited in a spreadsheet, and the evidence folder stores paths relative to the workspace so it can be committed to git.
The tool is aimed at teams that need a named person to answer for a release check, and at developers using an MCP-capable agent to test features on real devices. The project is MIT licensed and requires Node.js 22 or newer.
Tools this server provides (4)
Taken from the project documentation; the running server may differ.
- element_list
- Lists the names of elements that can be pressed.
- element_tap
- Taps an element by name.
- device_screen_text
- Reads text from the screen.
- device_screenshot
- Captures a screenshot of the device or page.
What it does
- Live AI-driven operation
- The AI follows each sheet step on the chosen device or page while the reviewer watches the live view.
- One-key verdicts
- D passes, F fails, A marks blocked, S skips, and Space leaves the AI verdict in place.
- Attributed evidence
- Each verdict is stored with the reviewer's handle, timestamp and the screen at that moment.
- Resumable runs
- Interrupted runs appear in a list and can be continued without re-entering earlier verdicts.
- MCP server access
- Agents can take screenshots, read screen text, list pressable names and tap, but cannot place verdicts.
- Local-only data
- Screens and evidence stay on the machine, with no data sent to outside services.
Good for
- Run a multi-platform regression sheet and review only the changed or failed cases
- Show an auditor who verified each test case and when
- Let an AI agent exercise a newly built feature on a real Android, web, desktop or iPhone target
- Run a sheet unattended overnight with --no-ui and review failures the next morning
Pricing
Open source
Free and MIT licensed; the installer is free and the source is on GitHub.
Prices change. Check meta-taro.github.io before you buy.
More servers for the same service
Git Qa FAQ
Does the AI ever give a final pass?
No. The AI can only record AUTO_PASS. A VERIFIED verdict is added only when a person presses the key, and it carries that person's handle.
What is needed to run it?
Node.js 22 or newer. adb is needed for Android, and a supported browser for web targets. ffmpeg and cwebp are optional.
Is test data sent to external services?
No. The README states that screens and evidence stay on the PC and the evidence folder contains no absolute paths or personal names.
Does it work on Windows?
Yes for web pages, Android devices and desktop apps. Recording needs ffmpeg, iPhone support on Windows is unverified, and the Windows installer is unsigned.
Is Git Qa free?
Yes. Git Qa is open source, so you can use and self-host it for free.
What are the best alternatives to Git Qa?
The closest alternatives to Git Qa are Avalyz, MCP Test Utils, Momentic, ego-webqa and UI Verify.
Where this comes from
- Description: written from the project's README, the project's website and its MCP Registry entry.