Playwright MCP Server – Browser Automation and Testing
Microsoft's official Playwright MCP server gives AI agents structured browser automation capabilities through Playwright. Use it to navigate websites, interact with page elements, inspect accessibility snapshots, capture evidence, automate browser workflows, and assist with web testing.
Overview
Microsoft's official Playwright MCP server connects compatible AI agents to a
real browser. It exposes browser automation through the Model Context Protocol
and returns structured accessibility snapshots that describe page elements,
roles, text, and references. This lets an agent interact with web pages without
requiring a vision model for routine browser control.
What the MCP server enables
Playwright MCP provides core browser tools and optional capability groups.
Depending on configuration, an AI agent can:
- Open pages, navigate URLs, go backward, and manage browser tabs.
- Read structured accessibility snapshots of the current page.
- Click buttons and links, type text, select options, and submit forms.
- Upload files, handle dialogs, and interact with supported page controls.
- Capture screenshots and inspect browser console messages or network activity.
- Save and restore browser storage state, including cookies and local storage.
- Connect to an existing Chrome or Edge session through the Playwright extension.
- Run in headed or headless mode and use Chromium, Firefox, or WebKit where supported.
- Enable additional tools through documented capabilities.
When to use it
Use Playwright MCP when an AI workflow must interact with a website or web
application. Typical uses include validating user journeys, reproducing UI bugs,
completing repetitive browser tasks, checking page content, testing forms,
generating evidence for QA, exploring an unfamiliar application, and assisting
with end-to-end test creation.
Connection and authentication
The standard configuration launches the local server over stdio with
npx @playwright/mcp@latest. Node.js 20 or newer is required, and the browser
downloads automatically on first use.
Playwright also supports a standalone local Streamable HTTP server. Launching
the package with --port 8931 exposes the endpoint at
http://localhost:8931/mcp. Neither local transport requires an API key.
Authentication to websites happens inside the controlled browser, and saved
browser state can persist between sessions.
Key considerations
Browser automation can access sensitive pages, session cookies, local storage,
downloads, and uploaded files. Use isolated profiles for untrusted tasks, avoid
exposing saved authentication state, and require human approval before purchases,
account changes, destructive actions, or submission of sensitive information.
Accessibility snapshots are the primary interaction model, while screenshots
are supplementary. For coding agents working primarily in large repositories,
Playwright also recommends considering its CLI and skills because they can be
more token-efficient than MCP in some workflows.
Supported Transports
stdio
Command: npx
Args:
-y@playwright/mcp@latest
streamable_http
URL: http://localhost:8931/mcp
Frequently Asked Questions
- When should an AI agent use the Playwright MCP server?
- Use it when an agent needs to interact with a website or web application, such as navigating pages, completing forms, reproducing UI bugs, validating user journeys, inspecting page content, or supporting browser-based testing.
- What does the Playwright MCP server add to an AI agent's capabilities?
- It gives the agent structured control of a real browser through accessibility snapshots and browser tools, allowing it to interact with live web pages instead of relying only on static HTML, screenshots, or manually supplied descriptions.
- What can an AI agent access or manage through Playwright MCP?
- Depending on enabled capabilities, the agent can navigate pages, inspect accessibility snapshots, interact with controls, manage tabs, upload files, handle dialogs, capture screenshots, inspect console or network activity, and save or restore browser state.
- How is authentication configured for the Playwright MCP server?
- The local MCP server itself does not require an API key. Website authentication occurs inside the browser, and login state can persist in a browser profile or saved storage-state file. Protect those files because they may contain sensitive cookies and session data.
- Which transport should be used for the Playwright MCP server?
- Use stdio with npx and @playwright/mcp@latest for the standard client configuration. Use the local Streamable HTTP endpoint at http://localhost:8931/mcp when the server must run separately from the MCP client, such as in worker or remote-display setups.