Playwright MCP Server

How to use

You run the Playwright MCP server locally and connect to it with an MCP client in your development environment. The server exposes a stream of structured page data and a set of actions you can perform, such as clicking elements, typing text, navigating pages, and capturing page snapshots. Use this to script web interactions, automate form filling, and extract data from structured content.

How to install

Prerequisites: You need Node.js and a package manager (npm or npx) installed on your machine. You should also have access to a terminal or command prompt.

# Install the Playwright MCP server via npx (runs the MCP directly without a global install)
npx @playwright/mcp@latest

Additional configuration and options

The Playwright MCP server can run in different modes and with specific transport settings. The canonical local setup uses a stdio transport. You can start the server with the following command, which makes it available for an MCP client to connect.

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": [
        "@playwright/mcp@latest"
      ]
    }
  }
}

Tools and capabilities

The Playwright MCP server exposes actions organized as snapshot-based interactions and vision-based interactions. You can perform a variety of tasks such as clicking, hovering, typing, selecting options, taking snapshots, navigating, and managing browser tabs. The primary actions include performing clicks, typing text, taking accessibility snapshots, capturing screenshots, and navigating through pages.

Notes on running modes and transports

Snapshot Mode uses accessibility snapshots for interaction, while Vision Mode uses screenshots. You can enable Vision Mode by adding a --vision flag when starting the server, though the default is Snapshot Mode for better performance and reliability.

Programmatic usage with custom transports

If you need to integrate with a custom transport, you can create a server instance and connect it to your transport layer, enabling you to deliver MCP messages over a custom channel.

browser_click

Perform a click on a web page element using the element description and the exact element reference from the page snapshot.

browser_hover

Hover the cursor over a targeted element using its description and the precise element reference from the snapshot.

browser_drag

Drag content from a start element to an end element using their descriptions and references.

browser_type

Type text into an editable element with options to submit or type slowly to trigger page handlers.

browser_select_option

Select one or more values in a dropdown by providing human-readable element descriptions and exact references.

browser_snapshot

Capture an accessibility snapshot of the current page to use for interactions.

browser_take_screenshot

Capture a screenshot of the current page for visual references when needed.

browser_tab_list

List all open browser tabs.

browser_tab_new

Open a new tab and optionally navigate to a URL.

browser_tab_select

Select a tab by its index.

browser_tab_close

Close a specific tab or the current one if no index is provided.

browser_navigate

Navigate the page to a specified URL.

browser_navigate_back

Go back to the previous page in history.

browser_navigate_forward

Go forward to the next page in history.

browser_console_messages

Return all console messages from the current page.

browser_file_upload

Upload one or multiple files to an input element.

browser_pdf_save

Save the current page as a PDF.

browser_wait

Wait for a specified number of seconds (up to 10 seconds).

browser_close

Close the current page or context.

browser_install

Install the configured browser if it is not already installed.