Most AI browser automation today runs through Chrome - either headless Chromium via Playwright or Chrome DevTools Protocol. A new open-source project called Safari MCP takes a different approach: native Safari control on macOS using AppleScript and JavaScript, with zero external browser dependencies.
The project exposes 80 tools through the Model Context Protocol (MCP is a standard that lets AI assistants like Claude Desktop or Cursor call external tools directly). Those tools span navigation, form filling, screenshots, cookie management, network capture, table extraction, and more. The practical appeal is straightforward: your AI agent operates inside your real Safari browser, with your existing cookies, logins, and sessions intact. No spinning up a separate browser instance. No re-authenticating.
How It Works Under the Hood
Safari MCP uses a dual-engine design. The primary path runs through a Safari extension that communicates over HTTP, executing commands in 5-20 milliseconds. When the extension isn't available, it falls back to AppleScript via a Swift daemon. Safari stays hidden in the background during automation - no window stealing focus while your agent works.
The project claims roughly 60% less CPU usage compared to Chrome-based automation on Apple Silicon, which makes sense given it's running through WebKit natively rather than spinning up a full Chromium process.
Setup requires Node.js 18+, enabling Safari's developer features, and allowing JavaScript from Apple Events. From there, you point your MCP client (Claude Desktop, Cursor, VS Code) at the server and the 80 tools become available to your AI agent.
What You Can Actually Do With It
The tool categories cover the basics you'd expect - clicking, scrolling, filling forms, taking screenshots, managing tabs - plus some less obvious capabilities:
- Network mocking and capture for testing how pages behave with different API responses
- CSS coverage analysis to find unused styles
- Geolocation overrides for testing location-dependent features
- Table and link extraction for structured data scraping
- Console log monitoring for debugging
The main limitation is that Chrome DevTools-exclusive features like Lighthouse performance audits aren't available, and pages with strict Content Security Policy or closed Shadow DOM elements need the extension path rather than the AppleScript fallback.
Who This Is For
This is a macOS-only tool, which immediately narrows the audience. But for Mac users who want their AI agents to automate tasks in Safari specifically - maybe because they need access to logged-in sessions, or they're building workflows that need to interact with sites that detect and block headless Chrome - it fills a gap that Chrome-based tools can't. The MIT license and straightforward MCP configuration mean you can try it in a few minutes if you already have an MCP-compatible AI client set up.