dattamks/ui-ux-visual-evaluator

This tool checks a website or app's design and usability automatically and produces one combined report covering visual design, ease of use, accessibility, how well it adapts to different screen sizes, and whether buttons and links actually work. It can look at a live website address or at uploaded screenshots, and it can compare two versions of a site against each other, or compare your site against a competitor's. What makes it different from a tool that just gives opinions is that it runs real automated checks behind the scenes. It injects a well known accessibility scanner called axe-core directly into the page to find actual violations of accepted accessibility standards, complete with the specific rule broken and where on the page it happened. It also measures things like how many different fonts, colors, and shapes are used across a design, which can reveal inconsistency, and it captures core web performance numbers like how fast the page first shows content and how much it visually shifts while loading. Beyond the automated measurements, the tool can also click through every clickable element on a page to check whether links and buttons actually work, reporting back which ones are broken, throw errors, or do nothing at all, complete with before and after screenshots. By default this clicking is safe and will not submit forms or delete anything unless you specifically allow it. On top of all these measured checks, the tool can optionally bring in an AI agent to add subjective judgment about usability and visual taste, reading the same evidence a human reviewer would. You can use it three ways: as a command line tool you run directly, as an MCP server that lets AI assistants call it directly, or as a packaged skill for Claude. Running it requires Python and an environment able to launch a real browser in the background, since that is how it captures live pages. It only reaches public web addresses unless you run it on the same network as a private site. Everything runs locally, so screenshots and findings stay on your machine unless you choose to share the report, and the project is released under the MIT license.