aryainguz/mrbrowser

Mr. Browser is a self-hosted tool for automating web browsers using plain English instead of technical CSS selectors. Rather than writing code that points to a specific button by its exact code name, which breaks the moment a website's layout changes, you describe what you want in ordinary terms, such as clicking a login button or typing into an email field, and the tool figures out which element on the page matches that description. The system works by understanding the structure of a page, scoring candidate elements based on your description, performing the action such as clicking or typing, then checking whether the action actually worked by comparing the page before and after. If a website's layout changes later, the tool can often recover automatically by matching a saved fingerprint of the element rather than a fixed selector, which is what the project calls self-healing. Workflows are written as YAML files, a simple, readable text format, listing a sequence of steps like opening a page, clicking something, typing text, or taking a screenshot. The project can run inside Docker for an easy setup, or be built locally if you have Go and Chromium installed. It also ships a Python library so you can control the browser automation directly from Python code, and it exposes a REST API so other programs can control it remotely over HTTP. The tool is described as using AI only optionally, with element matching handled mostly through algorithmic scoring rather than requiring a language model. Data such as saved element fingerprints is stored locally using SQLite. The project includes example workflows, a command line tool for running and debugging YAML workflows, and a built in step-through debug mode. It is released under the MIT license.