mvanhorn/agentcookie

What it solves

It eliminates the need to manually log into websites multiple times when running AI agents on a remote Linux server. Instead of manually copying cookies or performing "auth login" rituals for every site, it continuously synchronizes your active browser sessions from your local Mac to the remote machine where your agent operates.

How it works

agentcookie uses a source-and-sink architecture over a secure Tailscale network:

1. Source (Mac): Monitors the Chrome Cookies file for changes, decrypts them using the macOS Keychain, filters them based on a user-defined policy, and sends them encrypted (AES-256-GCM) to the sink.
2. Sink (Linux): Receives the encrypted cookies and injects them directly into the Chrome browser's in-memory store using the Chrome DevTools Protocol (CDP).

This allows any automation tool (like browser-use, Puppeteer, or Playwright) connecting to Chrome's debug port to immediately inherit your logged-in sessions without requiring a manual login process.

Who it’s for

Developers and AI researchers running browser-based agents on remote VMs, cloud runtimes, or homelab servers who want their agents to act on their behalf using existing authenticated sessions.

Highlights

• Continuous Sync: Automatically ships cookie diffs the moment they change on the source machine.
• Live CDP Injection: Injects cookies directly into memory on Linux, bypassing the need to rewrite Chrome's SQLite database.
• Secure Transport: Uses AES-256-GCM encryption over Tailscale's WireGuard channel.
• Flexible Policies: Supports both allowlist and blocklist modes to control which domains are synchronized.
• Multi-Profile Support: Auto-discovers and decrypts multiple Chrome profiles on macOS.