Safari 27 Adds Native MCP Server, Letting AI Agents Like Claude Code and Codex Control the Browser
Safari 27.0 introduces a local MCP server that lets AI coding agents control the browser to inspect DOM, network requests, and screenshots. Apple says the feature runs entirely on-device with no data sent to Apple.
Apple's WebKit team has detailed Safari 27.0, and the headline developer feature is a built-in Model Context Protocol (MCP) server that gives AI coding agents direct control over the browser window.
According to Apple's WebKit blog, the Safari MCP server lets tools like Claude Code, Codex, or "the agent of your choice" observe how code renders in real time. The server exposes access to the DOM, network requests, screenshots, and console output, letting an agent verify form states, checkout flows, and UI selections without a developer manually switching windows or pasting screenshots into a terminal.
Apple lists specific use cases for the feature, including comparing computed styles and layout against other browsers, testing accessibility issues such as missing labels, improper ARIA attributes, and poor contrast, and analyzing performance through navigation timing and resource load metrics.
Local-only, no Apple data collection
Apple states the MCP server runs entirely on the local machine, makes no network calls of its own, and has no access to a user's personal Safari data. Captured data goes directly to the agent in use rather than to Apple's servers, according to the company.
To enable it, developers must first turn on "Show features for web developers" under Safari > Settings > Advanced, which unlocks a Developer pane. From there, checking "Allow remote automation and external agents" activates the MCP server. Apple's blog post also includes the specific commands needed to connect Claude Code, Codex, or other MCP-compatible agents to Safari.
Broader Safari 27.0 changelog
Beyond MCP support, Apple says Safari 27.0 resolves nearly 850 issues — up sharply from the 525 fixes announced at WWDC just months earlier. Fixes span accessibility, CSS, encoding, fonts, forms, MathML, networking, and WebRTC, among other areas.
Other additions include support for immersive environments in visionOS, and an expansion of the HTML <model> element to iOS, iPadOS, and macOS, allowing websites to embed interactive 3D models the way they would images or video. A new Scroll Anchoring feature aims to stop pages from jumping when content like images loads above the user's current viewport position.
Web Inspector also gets upgrades, including inline color-contrast information, clearer redirect-chain visibility, and improved layout debugging tools. Apple's full post additionally covers changes to WebAssembly, WebGPU, JavaScript, CSS, WebRTC, media playback, Web Extensions, WKWebView, and Service Workers.
What this means
Safari becomes the first major browser to ship a first-party MCP server directly in the shipping product rather than through a third-party extension, positioning Apple to compete for the growing AI coding-agent workflow where Chrome DevTools and third-party MCP bridges have dominated. By keeping the server local and network-call-free, Apple is drawing an explicit privacy line against cloud-based browser-automation tools. The real test will be developer adoption: MCP support is only useful if agent makers like Anthropic and OpenAI's tooling ecosystems treat Safari as a first-class target rather than an afterthought behind Chromium-based browsers.
Related Articles
Anthropic Rebuilds Claude Code Projects to Run Parallel AI Agent Threads
Anthropic has rebuilt Claude Code's Projects feature so a coordinator agent splits user goals into parallel cloud-based threads, each capable of opening pull requests and running tests. The beta is limited to select Pro and Max subscribers, with Team, Enterprise, and local execution support coming later.
Anthropic Relaunches Claude Code Projects to Coordinate Multiple Cloud Agents
Anthropic has relaunched Projects in Claude Code, letting users direct multiple AI agent 'threads' that work in parallel under a coordinating agent. The beta feature rolls out today to select Claude Pro and Max subscribers, with broader access planned later.
iOS 27's Rebuilt Siri AI Handles 125 Real-World Tasks in Independent Test
Journalist David Pogue ran 125 test prompts against the new Siri AI in iOS 27, finding it successfully handled cross-app queries, contextual follow-ups, and complex multi-step requests. Failures were limited to tasks requiring third-party app updates, such as Venmo payments.
GitHub Used Copilot to Rewrite Its Own Copilot Runtime in 800,000 Lines of Rust
GitHub used its own Copilot agent to help migrate the Copilot runtime to Rust, producing roughly 800,000 lines of production code. The company says the scale of the rewrite would not have been feasible without agentic coding assistance.
Comments
Loading...