javimosch Playwright MCP — Browser Automation for AI Agents
Official Microsoft MCP server (32.5k⭐). Control browser via MCP tools over stdio. Works with any MCP client.
Quick Start
sc playwright-mcp self mcp # Register MCP server
# Then agents can use browser tools directly
MCP Tools Available
Once registered, agents get these tools:
| Tool | Description |
|---|---|
browser_navigate |
Navigate to a URL |
browser_click |
Click element by selector |
browser_fill |
Fill input field |
browser_snapshot |
Get page state (accessible snapshot) |
browser_screenshot |
Take screenshot |
browser_press_key |
Press keyboard key |
browser_close |
Close current page |
Usage (via Agent)
Agents can say:
- "Navigate to https://example.com"
- "Take a screenshot of the page"
- "Click the login button"
- "Fill the search box with 'query'"
Requirements
- Node.js 18+ (for npx)
- First run auto-downloads @playwright/mcp from npm
Notes
- Each agent session spawns a new browser context
- Screenshots return base64-encoded PNG
- Uses Chromium by default (auto-installed)
- For persistent sessions, keep the daemon running