Playwright MCP on GitHub: what it does and how to use it with AI agents

What Playwright MCP does

Gives your agent a browser it can control through Playwright. It works from the page accessibility tree instead of screenshots, so the agent can click, type and read pages to test your app or fill in a flow.

Who it is for

Builders who want their coding agent to check the app it just changed, and anyone automating a web task.

How to use Playwright MCP with Claude Code, Codex or Cursor

1. Claude Code: claude mcp add playwright npx @playwright/mcp@latest
2. Cursor or Codex: add npx @playwright/mcp@latest as an MCP server.
3. Ask: "Open localhost:3000, sign up with a test account and tell me what breaks."
4. Run it headless in CI with the flags listed in the README.
5. Since v0.0.84 (Oct 31, 2026) you can keep two logins apart: open a tab with browser_tabs, action new and isolatedContext set to a name, and tabs with different names do not share cookies or storage.

Tip: Use a separate browser profile. Do not let an agent browse while logged in to accounts you care about. If pages put tokens in the URL, start it with --redact-url-params=access_token,code so those values show as <redacted> in tool output. The release notes call this a convenience, not a security feature.

Related repos

• Browser Use: Open-source Python library for AI agents that use a web browser.
• MCP Servers: The official reference servers for the Model Context Protocol.

Playwright MCP FAQ