Hacker-Valley-Media/Interceptor

What it solves

Interceptor provides a way for AI agents to control a user's actual, live browser and macOS applications as if they were a human. It eliminates the need for separate automated browser instances (like those used by Playwright or Puppeteer) which often lose authenticated sessions, cookies, and logins, and are easily detected by websites.

How it works

Interceptor operates through a CLI and a daemon that bridges the gap between an AI agent and the OS. It offers three primary "surfaces":

• Browser: A WebExtension for Chrome, Brave, and Safari that allows agents to read DOM trees, click, type, and observe network traffic within the user's existing session.
• macOS: A Swift bridge daemon that uses accessibility trees and OS-level trusted input to drive native desktop applications.
• iOS: An on-device XCUITest runner that allows agents to control apps on an unlocked iPhone via WiFi.

Agents interact with the project via CLI commands (e.g., interceptor act e1) or as an MCP (Model Context Protocol) server, which integrates directly with AI clients like Claude Desktop or Cursor.

Who it’s for

• AI agent developers building "computer-use" or "browser-use" capabilities.
• Users who want AI agents to automate tasks within their already-logged-in web and desktop environments.
• Developers needing to automate canvas-rendered editors (like Google Docs) or capture native client-side exports without manual save dialogs.

Highlights

• Session Persistence: Uses the user's real browser profile, keeping cookies and logins intact.
• Cross-Platform Control: Unified CLI for web, native macOS apps, and iOS apps.
• MCP Integration: Native support for MCP-native AI clients for easy tool registration.
• Advanced Browser Interaction: Bypasses isTrusted checks and handles canvas-rendered editors without needing OS-level keyboard events.
• Passive Network Monitoring: Captures fetch, XHR, and WebSocket traffic without triggering debugger banners.