Back to directory

Claude for Safari

A skill that lets an AI agent read, control and take screenshots of a user's own Safari browser on macOS through AppleScript, without a browser extension or a separate automated browser instance.

Claude for Safari

Claude for Safari gives an AI agent, in particular Claude Code, the ability to control the user's existing Safari browser on macOS. It works by using macOS's own automation and AppleScript facilities rather than a browser extension, a proxy, or launching a separate browser session, so it does not hijack an existing logged-in session the way tools such as Playwright can.

Capabilities include listing open windows and tabs with their titles and addresses, extracting page text and structured or simplified page content, running arbitrary JavaScript in the page, taking a screenshot of the Safari window, navigating to addresses and opening tabs or windows, and clicking elements built with frameworks such as React, Vue or Angular. It installs as a single command that adds the skill, and is compatible with any AI agent that supports the shared skills format, including Cursor and Windsurf.

First-time setup requires granting the terminal or agent host permission to automate Safari in macOS's privacy settings.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.