The MCP server, built for operatives, provides an in-editor, browser-driven autonomous agent workflow to test, debug, and fix web applications. It exposes tools like web_eval_agent, an automated UX evaluator that drives a real browser to navigate, test flows, capture screenshots, and collect console and network logs, producing a comprehensive UX report. It also includes setup_browser_state, which opens a non-headless browser session for sign-in and authentication, reusing cookies and local storage across runs. Users interact with the MCP server through chat-style prompts in their IDE, for example evaluating an app at http://localhost:3000 with a task description such as run through the signup flow and note UX issues. The system integrates with Playwright, uv, and Cursor to enable end-to-end validation: the Cursor agent calls the web QA agent to verify that code changes work as expected. Outputs include a structured report, console logs, network activity, and a timeline of events, helping developers identify UX issues quickly and iteratively improve software.