Automation, extensions, and devtools

Driving the browser from a terminal, an agent, or Siri

Automation is off by default, and it unlocks in layers. Reading pages, running scripts, taking screenshots, and sending genuine mouse clicks are four separate switches, and real clicks additionally need macOS Accessibility permission.

Things to try

about 7 minutes

01In Settings, Security, turn on agent automation, then enable only page reading.
02Link the CLI that ships inside the app bundle.sudo ln -sf /Applications/Browser.app/Contents/Helpers/browser-cli /usr/local/bin/browser-cli
03Open a page from the terminal and read it back.browser-cli open https://example.com && browser-cli wait, then browser-cli snapshot. Commands travel through an owner-only named pipe.
04Send yourself a message in the real browser window.browser-cli notify "Please solve the captcha". This is the whole point: the automation can hand a login or a captcha back to a human and pick up afterward.
05Connect an AI tool that speaks MCP.browser-cli install-mcp all wires up Codex and Claude Code if they are installed. 53 BrowserOS-compatible tools, stable per-window page handles, and parallel hidden pages. It uses the same in-app switches, and its events join the same local timeline.
06Open the Shortcuts app and look for Browser's actions. Then ask Siri for a new tab.Open URL, Search the Web, and New Tab show up in Shortcuts, Spotlight, and Siri.
07From the Extensions menu, load an unpacked extension folder and approve its scopes.⌥⌘EOpen its popup, control whether it sees private tabs, and remove it from the management panel.
08Open developer tools, the console, and the element picker.⌥⌘I⌥⌘J⌥⌘CSettings, Appearance docks devtools to the bottom, left, or right, or keeps them in their own window.

Worth filming

The captcha handoff. A script stops, a human solves it in the window that was there all along, the script continues.

Settings panes behind this chapter: Security.