Apple Adds an Official MCP Server for Testing Sites in Safari

Apple has released a Safari MCP Server in Safari Technology Preview 247, giving web developers an official way to connect AI coding agents directly to WebKit for page testing, inspection, and debugging.

The tool implements Model Context Protocol support inside Apple's experimental browser.

Once an MCP compatible client is connected, an AI agent can open pages in Safari Technology Preview, read the page structure, pull console output, review network activity, capture screenshots, and check whether what renders matches what the developer expected.

The feature is aimed at developers rather than everyday browsing. Apple's WebKit team describes it as a browser compatibility tool built to make web development and debugging faster.

What the server actually does

The Safari MCP Server is tied to safaridriver with MCP support, which gives Safari a native place inside AI coding workflows for the first time.

Until now, most coding assistants worked comfortably inside editors, terminals, and Chromium based automation, while Safari usually required an extra step. That gap mattered most when testing WebKit specific layout, styling, or mobile behavior. The new server folds that testing into the same development loop.

Through the connection, an agent can load a site, inspect computed styles, check layout, read page content, review debugging information, and use screenshots to compare the rendered result against the intended design.

Apple frames this as a way to find and fix issues, not as a general AI browsing feature.

Testing only in Chrome or a Chromium based tool can miss WebKit behavior. Layout, forms, media playback, scrolling, input handling, typography, and CSS support can all render differently, and iPhone specific rendering is a common source of surprises.

That is the case Apple is making for the server. Safari is the default browser on iPhone, iPad, and Mac, and iPhone traffic is too large for production teams to treat WebKit as a secondary target. A rendering difference that seems small can still break navigation, checkout, sign in, media playback, or accessibility.

Giving an AI agent direct access to a real Safari environment is meant to surface those problems earlier, ideally while code is still being written rather than during a late cleanup pass.

How MCP fits into AI coding

Model Context Protocol is a standard for connecting AI assistants to outside tools and data. An MCP server exposes functions that an agent can call during a task, so instead of only reading files, the agent can interact with a browser, inspect logs, or collect the state of a running app.

Apple's implementation applies that model to browser debugging. A developer can ask an assistant why a layout breaks in Safari, and the agent can inspect the rendered page rather than guessing from source code alone. Computed styles, DOM structure, console warnings, failed requests, and screenshots give the model the kind of evidence that plain code review often cannot.

The approach brings Apple closer to the browser agent setups already available through Chromium tools, Playwright workflows, and third party MCP servers. The difference is that WebKit now has an official bridge inside Apple's preview browser.

Setup and requirements

Developers need Safari Technology Preview 247 on macOS to try the feature. Apple says the server can be enabled through Safari's developer tools path for remote automation and external agents, after which an MCP compatible client connects to safaridriver's MCP mode.

Safari Technology Preview is Apple's experimental channel for upcoming WebKit changes, so its behavior can differ from stable Safari. That makes it useful for early compatibility work, but production bugs should still be confirmed against the public release of Safari when it matters.

It is also not a replacement for full QA. An agent can identify likely causes, test changes, and surface browser evidence, but it can miss edge cases. Human review stays necessary for accessibility, visual polish, payment flows, privacy sensitive pages, and anything tied to user data.

A more official path than community tools

Developers were already experimenting with Safari automation before this release. Some projects used AppleScript to drive the browser, while others combined Safari extensions, Swift tools, or WebKit debugging access to expose controls to AI clients. Those efforts showed real demand for Safari sessions, logins, and WebKit behavior inside coding agents without moving every test to Chrome.

Apple's version is narrower and more focused on inspection and debugging, and that restraint works in its favor.

By keeping the first release inside Technology Preview and tied to developer automation, Apple can test the model with a technical audience before deciding how much of it belongs in stable Safari.

Browser agents raise security and privacy questions once they can control sessions or read content inside logged in sites, so a developer only starting point is the cautious one.

The Safari MCP Server is still a preview feature. The real test will be whether developers start catching Safari bugs earlier, before layout problems, console errors, or mobile rendering issues reach production.