Safari Ships MCP Server That Lets AI Agents Inspect and Debug Web Pages
Safari Technology Preview 247 bundles an MCP server with 17 tools that let AI coding agents inspect the DOM, capture screenshots, read console output, and run JavaScript against a live browser, all running locally with no data leaving your machine.
WebKit shipped an MCP server inside Safari Technology Preview 247 on July 1, giving AI coding agents direct access to a live Safari browser window. Your agent can now see how code actually renders rather than guessing from source alone.
The server exposes 17 tools: browser_console_messages for log access, evaluate_javascript to run code on the page, screenshot for visual capture, page_interactions for click, type, scroll, and hover, list_network_requests for request summaries, and get_page_content to extract text as markdown, HTML, or JSON. There is also viewport control, tab management, and media emulation for testing print stylesheets. Setup is a single CLI command for Claude or Codex after enabling remote automation in Safari settings.
Everything runs locally. The server makes no network calls and does not touch personal Safari data like AutoFill or passwords. It uses the existing safaridriver binary that ships with Safari Technology Preview, so there is nothing extra to install.
For frontend developers shipping to multiple browsers, this changes your debugging workflow. Instead of manually reproducing a layout bug in Safari to grab a screenshot and console output, you describe the problem to your agent and it pulls the evidence directly. No window hopping, no manual inspector diving. The WebKit team positioned this as a first class tool for accessibility checks, performance profiling, and verifying UI state without leaving your terminal. If you work on a Mac and already use an AI coding agent, this is worth the five minute setup.