Playwright MCP
An open-source browser automation server for AI agents that runs locally on macOS, Windows or Linux and reads page structure without a vision model.
Playwright MCP lets AI agents control a browser by reading structured page information rather than interpreting screenshots. It's for developers building browser automation, exploratory tests or agent workflows that need to keep a browser session open across repeated actions. The server runs locally on macOS, Windows and Linux, or as a self-hosted service.
Its main distinction is how agents see pages: Playwright's accessibility tree gives the model identifiable elements to act on, reducing the ambiguity of choosing targets from images. It doesn't require a vision model. Agents can click elements, drag and drop, upload files or data, and inspect browser console messages.
It connects to MCP clients including LM Studio, Cursor, VS Code, Claude Desktop and Codex. Browser support includes Chrome, Firefox, WebKit and Edge, with visible or headless operation and mobile device emulation. Docker deployment supports headless Chromium only.
Persistent profiles retain login information locally between sessions, with separate profiles for different projects. Isolated sessions keep the profile in memory and discard its state when the browser closes. A browser extension can also connect the agent to an existing Chrome or Edge session.
The project is open source under Apache 2.0 and requires Node.js. Its MCP interface suits workflows that need continuous browser context and repeated inspection of page structure; Playwright CLI with skills uses fewer model tokens for coding-agent workflows.
