Playwright MCP MCP server

Playwright MCP is Microsoft's Model Context Protocol server for browser automation. It gives an agent a browser it can open, click through and read, and it describes each page as a structured accessibility snapshot rather than a picture, so the model needs no vision support to act on what it sees. Setup needs Node.js 18 or newer. The standard config most clients accept is an entry named playwright whose command is npx with the argument @playwright/mcp@latest. In Claude Code the README gives a single line: claude mcp add playwright npx @playwright/mcp@latest We think it earns its place when an agent has just changed your app and you want it to check the result itself: load the page, fill in the sign-up form, press the button and report what the page actually says. By default the browser keeps a persistent profile, so a login survives between sessions, and an --isolated flag keeps the profile in memory for clean test runs. Two cautions from the README. It states plainly that Playwright MCP is not a security boundary, so it should not be treated as a sandbox for untrusted sites. And the maintainers now point coding agents toward their Playwright CLI with skills instead, because loading tool schemas and accessibility trees into the context costs tokens; they keep MCP for long, stateful sessions where that cost is worth it.