WebKit has introduced a Safari MCP server in Safari Technology Preview 247, giving coding agents a direct way to look at what a web page is doing inside a Safari browser window.
Saron Yitbarek, writing for WebKit, said the server implements the Model Context Protocol and can be used by any MCP-compatible client. The practical effect is straightforward: instead of a developer describing a broken layout, console error, or Safari-only bug to an agent, the agent can inspect the running page itself.
That changes the debugging loop for developers who already use agents in their editor or terminal. The agent can see page content, query the DOM, read console output, inspect network requests, take screenshots, and run JavaScript in the page. It can also interact with the page by clicking, typing, scrolling, hovering, and sending key presses, according to WebKit.
What the server exposes
WebKit lists a set of browser-control and inspection tools behind the MCP server. They cover tab management, page loading, screenshots, JavaScript evaluation, network request summaries and details, browser dialogs, page metadata, viewport sizing, media emulation, page interactions, and waiting for navigation to finish.
For Safari-specific testing, WebKit says an agent can open a site in Safari, inspect computed styles, check layout, and compare the page state against what the developer expects. For performance work, the agent can evaluate JavaScript on the page and surface metrics such as navigation timing and resource load times.
WebKit also pitches accessibility checks as a use case. The server gives an agent enough page access to look for common issues such as missing labels, incorrect ARIA attributes, and poor contrast, according to Yitbarek. Those are still agent-driven checks, so developers should treat the results as useful triage rather than a replacement for proper testing.
The server can also verify page state. WebKit says an agent can inspect a form, query elements with selectors, confirm interactions, and check different states in flows such as checkout. In plain terms, Safari is giving the agent a controlled periscope into the browser instead of relying on the developer to narrate what went wrong.
Setup and privacy claims
To use the feature, developers need Safari Technology Preview installed. WebKit says they must enable “Show features for web developers” under Safari Settings, Advanced, then enable “remote automation and external agents” under Safari Settings, Developer.
For Claude and Codex, WebKit provides terminal commands that add the Safari Technology Preview safaridriver binary as an MCP server with the “--mcp” argument. For other agents, developers can add an entry to an MCP or config JSON file pointing to “/Applications/Safari Technology Preview.app/Contents/MacOS/safaridriver” with the same argument. WebKit notes the server name is arbitrary.
Yitbarek wrote that the Safari MCP server runs on the user’s local machine and makes no network calls of its own. WebKit also says it does not have access to personal information in Safari, including AutoFill or other browser activity.
That does not mean captured data stays private from the tools a developer chooses. WebKit says page content, screenshots, and console logs go to the agent being used, not to Apple. What happens after that depends on the agent and model, so the boring security advice applies: do not hand browser access to an agent you do not trust.
WebKit is asking developers who hit problems to file reports through its bug tracker.
This story draws on original reporting from WebKit.
