Microsoft Playwright MCP

Microsoft's official MCP server for deterministic browser automation via Playwright's accessibility tree.

MCP ServerMCP-Compatible AgentsOfficialVerified

View source Documentation

Category: Test Automation
Author: Microsoft
Pricing: free
License: Apache-2.0

Connect Microsoft Playwright MCP to your agent

claude mcp add --transport stdio playwright -- npx @playwright/mcp@latest

Add the server
Run this command in your terminal.
Finish setup
Claude Code starts this server locally. Restart Claude Code and retry if the server does not appear.
Try it
Ask your agent: “Open example.com and return the page title.”

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

Add the server
Add this configuration to Cursor's user-level or project-level mcp.json.
Finish setup
Enable the server in Cursor's MCP settings. Restart Cursor and retry if the server does not appear.
Try it
Ask your agent: “Open example.com and return the page title.”

codex mcp add playwright -- npx @playwright/mcp@latest

Add the server
Run this command in your terminal.
Finish setup
Codex starts this server locally. Restart Codex and retry if the server does not appear.
Try it
Ask your agent: “Open example.com and return the page title.”

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

Add the server
Add this configuration to your client's MCP configuration file.
Finish setup
Save the configuration and restart your client if the server does not appear.
Try it
Ask your agent: “Open example.com and return the page title.”

What it does

Microsoft Playwright MCP is the official MCP server that provides browser automation via Playwright's structured accessibility tree, rather than pixel or screenshot analysis, so it needs no vision model and stays deterministic and reliable for test automation. It exposes roughly 68 tools, including assertion tools — browser_verify_element_visible, browser_verify_text_visible, browser_verify_list_visible, browser_verify_value — available opt-in via --caps=testing, along with browser_snapshot, browser_console_messages, browser_network_requests, and DevTools/recording tools for video and tracing.

How QA teams use it

Give any MCP-capable AI assistant a deterministic browser for driving E2E test scenarios without relying on screenshots or vision models.
Enable the testing capability flag to get direct assertion tools for visibility, text, list, and value checks.
Pull console messages and network requests during a debugging session, and capture trace/video artifacts for failing runs.