Playwright MCP
Microsoft's official MCP server for driving a browser with Playwright through the accessibility tree, no screenshots needed
- 38,768 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The agent opens arbitrary sites and can submit forms
- The persistent profile keeps cookies and sessions
Install
- Works in
Manual install
claude mcp add playwright npx @playwright/mcp@latestAccess and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent opens arbitrary sites and can submit forms
- The persistent profile keeps cookies and sessions
This is third-party code. Review the repository files before installing.
What it does
The server launches a browser through Playwright and gives the agent tools to navigate, click, type and read pages. Pages reach the model as structured accessibility tree snapshots, so no vision model is needed. You can pick Chrome, Firefox, WebKit or Edge, emulate a mobile device, run headless, or attach to an open browser through an extension. Extra vision, pdf and devtools capabilities can be switched on.
Who it is for. For developers and QA engineers who need their agent to work with live web pages.
Good fit when
- The agent must walk through a site flow: log in, fill a form, check the result
- You need exploratory automation or self-healing tests with a persistent browser context
- You need data from a page that renders on the client
Not a fit when
- A coding agent in a large project needs to save context: the README recommends Playwright CLI with skills
- A plain HTTP request without a browser is enough
Example request
Open our staging site, register a test user and check that the confirmation email notice appearsMCP
- Transport
- stdio
- Authentication
- not required
Environment variables
PLAYWRIGHT_MCP_BROWSER- Browser or Chrome channel: chrome, firefox, webkit, msedge
PLAYWRIGHT_MCP_HEADLESS- Run the browser headless
PLAYWRIGHT_MCP_CAPS- Extra capabilities: vision, pdf, devtools
PLAYWRIGHT_MCP_SECRETS_FILEsecret- Path to a dotenv file with secrets
Limitations
Requires Node.js 18 or newer. Page snapshots and tool schemas use many tokens. A persistent browser profile can serve only one client at a time; parallel clients need --isolated or a separate --user-data-dir. Allowed and blocked origin lists are not a security boundary.
How to disable. Remove the playwright server from your client's MCP config, in Claude Code with claude mcp remove playwright.
FAQ
How does MCP differ from Playwright CLI?
The README recommends the CLI with skills for coding agents because it uses fewer tokens. MCP fits cases that need a persistent browser and rich page introspection.
Can I keep logins?
Yes, a persistent profile is used by default. For isolated sessions pass a storage state file with --storage-state.
How is Playwright MCP different from screenshot-based browser tools?
The server hands the model the page as a structured accessibility tree snapshot, not an image. The agent clicks and types by element, which is more precise and cheaper than screenshots. The project is official, from Microsoft.
How do I connect Playwright MCP to Claude Code and Cursor?
In Claude Code: claude mcp add playwright npx @playwright/mcp@latest. The server also works in Cursor, VS Code, Claude Desktop, Codex, Windsurf, Goose and Amp, with a config for each in the install block.
How is the official Playwright MCP different from community servers?
It is the server from the Microsoft team itself: it hands the page to the model as an accessibility tree, and browsers, modes and flags evolve together with Playwright. Community servers such as ExecuteAutomation and Automata Labs ship their own tool sets, and some generate test code.
Discussion
No questions or reviews yet.
Related
MCP serversMCP Server Playwright by Automata Labs
MCP Server Playwright
Playwright-based MCP server for browsing sites, clicking, filling forms, taking screenshots and running JavaScript in a browser
An MCP server and Chrome extension that let the agent drive your already open browser with your logged-in sessions, all locally
An MCP server that lets an agent drive a real browser: open pages, click and fill forms through Playwright
An MCP and CLI that let an agent control your real Chrome, with its own logins and extensions, through Playwright
A Chrome extension with an MCP server: the agent drives your everyday browser with its sessions, tabs, bookmarks and history
An MCP server that opens pages in headless Chromium through Playwright and hands the agent the main text as Markdown, even from JavaScript-heavy sites
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Playwright MCP (this entry) | yes | not required | 38,768 | |
| Playwright MCP Server | no | not required | 5,675 | generates test code, emulates devices |
| MCP Server Playwright by Automata Labs | no | not required | 297 | third-party server on an early SDK |
README in short
The README first compares MCP with Playwright CLI and explains when to choose each. It then gives a standard JSON snippet and install commands for VS Code, Cursor, Claude Code, Codex, Goose, Windsurf and other clients. A large section lists launch arguments with matching environment variables, plus notes on user profiles, initial state and the config file. Apache-2.0 licensed.