Playwright MCP

Playwright MCP is Microsoft's Model Context Protocol server for browser automation, built on the Playwright framework. Its defining design choice is that it drives pages through the browser's accessibility tree rather than screenshots, so the agent reads a structured snapshot of elements, roles, and labels and acts on stable references instead of guessing pixel coordinates. That makes it fast, deterministic, and usable by models without vision, though a visual mode is available when a screenshot is genuinely the right tool. The tool set covers a full session: navigating to a URL, going back and forward, clicking, typing, hovering, selecting from dropdowns, dragging, pressing keys, uploading files, handling dialogs, managing multiple tabs, waiting on text or elapsed time, and resizing the viewport. For debugging it can read console messages and captured network requests, and it can save a page as PDF or take screenshots. It runs against Chromium, Firefox, or WebKit, headed or headless, and can attach to a persistent profile so the agent works inside an already authenticated session. It is a common way to give a coding agent hands on a real browser, whether that means reproducing a bug report, verifying a UI change end to end, filling a form on a site that has no API, or reading content that only exists after JavaScript runs. Give it a scoped browser profile and remember that any site it can reach is a site the agent can act on.