Safari MCP

Executes browser control commands directly against Safari's native API layer using AppleScript, bypassing Chromium/WebDriver overhead. Leverages macOS Automation framework to send typed commands (navigate, click, type, scroll) to Safari's runtime without spawning separate processes or maintaining WebDriver connections. Maintains persistent Safari session state including authentication cookies and login sessions across command sequences.

Exposes 80 Safari automation operations as discrete MCP tools (resources/functions) that AI agents can discover and invoke through the Model Context Protocol. Each tool maps to a specific Safari automation capability (navigate, click, extract text, take screenshot, etc.) with typed parameters, return schemas, and error handling. Implements MCP server interface to advertise tool availability, handle invocation requests, and stream results back to client agents.

Detects and reports automation errors including navigation failures, element not found errors, timeout conditions, and JavaScript execution errors. Implements page state validation to check for expected elements, error messages, or page conditions before proceeding. Returns structured error objects with error codes, messages, and context information. Supports conditional error handling in agent workflows.

Executes Safari automation commands without displaying browser windows or UI elements, using AppleScript's background execution mode and Safari's headless-compatible operation flags. Commands run in Safari's background process without visual feedback, allowing agents to perform browser operations without interrupting user workflows or consuming screen real estate. Maintains full automation capability (navigation, interaction, data extraction) while remaining invisible to the user.

Maintains Safari's native cookie store, session tokens, and authentication state across automation command sequences by operating against the same Safari process instance. Automation commands execute within the context of the user's existing Safari session, automatically inheriting all stored credentials, cookies, and login state without explicit session management or cookie jar manipulation. Leverages macOS Keychain integration for secure credential storage when needed.

Extracts structured content from loaded Safari pages using CSS selectors, XPath expressions, and JavaScript evaluation. Supports querying page DOM to retrieve element text, attributes, computed styles, and nested structures. Returns extracted data as structured JSON objects with optional HTML sanitization. Implements efficient DOM traversal using Safari's native JavaScript engine rather than separate parsing libraries.

Controls Safari's navigation stack by loading URLs, handling redirects, managing browser history (back/forward), and detecting page load completion. Supports both direct URL navigation and relative link following. Implements load-state detection to wait for pages to fully render before returning control to agent, preventing race conditions in multi-step workflows. Handles redirect chains transparently.

Simulates user interactions with page elements including mouse clicks, keyboard input, and scrolling operations. Supports clicking by CSS selector or coordinates, typing text into form fields with optional key modulation (shift, ctrl, etc.), and scrolling to specific elements or coordinates. Implements element visibility checking before interaction to prevent clicking hidden elements. Uses AppleScript event simulation for native input handling.

Captures full-page or viewport screenshots of Safari pages in PNG or JPEG format, with optional element-specific cropping. Screenshots include rendered CSS, JavaScript-generated content, and current page state. Supports both full-page captures (including off-screen content via scrolling) and viewport-only captures. Returns image data as base64-encoded strings or file paths for integration with vision models or visual validation workflows.

Executes arbitrary JavaScript code within Safari's page context, with access to the DOM, window object, and page-loaded libraries. Returns JavaScript evaluation results as JSON-serializable values. Supports both synchronous execution and promise-based async operations. Enables complex page interactions and data extraction that cannot be expressed through declarative selectors. Runs code in the page's global scope with full access to page state.

Controls Safari's tab and window management including creating new tabs, switching between tabs, closing tabs, and managing multiple windows. Supports tab enumeration to discover open tabs and their URLs/titles. Implements tab context switching to execute subsequent commands in specific tabs. Enables workflows that require parallel operations across multiple tabs or windows.