What is munk-ai?

Munk AI is a testing tool that watches apps run on real devices and browsers, then reports what it sees back to AI coding tools so they can fix problems without a human in the loop. The central problem it addresses is that AI tools are now quite good at writing code, but after the code is written, someone still has to open the app, tap through screens, check whether things work, take screenshots of failures, and describe what went wrong. Munk AI automates that verification step. When a developer or an AI coding agent makes a change to an Android, iOS, or web app, Munk AI can take a plain-language description of what the app should do, convert it into a test plan, execute that plan on a real device or browser, and produce structured output: screenshots, UI element trees, and runtime logs. If something fails, that evidence goes back into the coding agent's context so it can attempt a fix, creating a closed loop that does not require a human to click through each build. The tool runs locally on macOS (with Windows and Linux support on the roadmap). Installation is a single shell command, and it starts a local web interface for managing tests, recordings, and results. It also exposes an API and supports MCP, a protocol that lets coding assistants call tools directly, so it can be connected to AI development environments without extra configuration. Under the hood the core runtime is Python, using FastAPI and a computer-vision library called OpenCV to analyze what appears on screen visually rather than relying on fragile element selectors. Android testing uses uiautomator2, web testing uses Playwright with a Chromium browser, and the local web UI is built with Vue 3 and TypeScript. The project is under active development. The repository is public but core modules are being opened in stages. It is licensed under AGPL-3.0.