actionbook/actionbook

Let your AI agent get the sources behind logins and paywalls.

What it solves

Actionbook addresses the inefficiency and fragility of traditional AI browser agents. Most agents are slow because they must parse a page snapshot after every action, struggle with modern web elements like virtual DOMs and login walls, and can only process one page at a time. Actionbook enables agents to operate websites faster, more accurately, and concurrently.

How it works

It functions as a remote Model Context Protocol (MCP) connector paired with a Chrome extension. This setup allows an AI agent to reuse the user's active logged-in browser sessions. The system uses "action manuals" to give agents precise instructions on how to interact with a site, utilizing direct API calls when possible and falling back to UI automation when necessary. This stateless architecture allows the agent to operate dozens of tabs in parallel.

Who it’s for

Users who want to automate complex web-based workflows—such as gathering data from multiple company websites or creating reports from logged-in accounts—using AI agents in their browser.

Highlights

  • Concurrent Operation: Can operate multiple tabs in parallel to perform tasks across many websites quickly.
  • MCP Integration: Connects directly to AI apps like Claude, ChatGPT, and Perplexity via a hosted MCP endpoint.
  • Login Persistence: Uses a Chrome extension to operate within the user's existing authenticated sessions.
  • Recipe System: Allows users to save successful tasks as "Recipes" to be rerun with new inputs.

Related

  • Project
  • Project
  • Project
  • Project
  • Project