Actionbook lets an AI agent reach data behind logins and paywalls by pairing a hosted MCP connector with a Chrome extension that drives your real, already-signed-in browser.

What it is

Actionbook is an AI agent that works inside your browser: it reads data behind your logins, operates the websites you already use, and returns finished work such as a sourced report, a clean CSV, or ready-to-send drafts rather than raw answers. It lives in the Rust ecosystem under an Apache-2.0 licence and ships as a remote MCP connector (https://edge.actionbook.app/mcp) paired with the Actionbook Chrome extension, so no CLI install is required.

The concrete problem it solves is that ordinary agents are slow, brittle, and serial on modern websites. Because an agent normally takes a snapshot after every step, parses the page, and then decides what to do next, searching one room on Airbnb takes five minutes. Virtual DOMs, streaming components, and single-page apps defeat agents on dropdowns, date pickers, and login walls, and an agent finishes one page before it can start the next — thirty company websites means thirty rounds, one after another. Actionbook replaces this snapshot-and-guess loop with action manuals that tell agents exactly what to do, operating the sites you work in every day through direct API requests where possible and UI automation where not, with login handled either way.

Key capabilities

• A single actionbook MCP tool whose subcommands cover action lookup (search, manual) and browser control (goto, snapshot, click, type), with the full command set discovered from the tool itself.
• Action manuals that tell agents exactly what to do, removing the parse-page-then-decide step the README describes as the source of slowness.
• Direct API calls when possible with UI automation as fallback, so login walls are handled on either path.
• Reuse of your real logged-in Chrome sessions, because the extension drives your actual browser rather than a separate isolated one.
• A stateless architecture that operates dozens of tabs in parallel — the README shows an agent visiting 192 First Round portfolio company websites and collecting their taglines in 2 minutes.
• Recipes: saving a task that works the way you want and running it again with new inputs.
• An agent in the Chrome side panel that works beside you on the current page and delivers outputs such as sourced reports, clean CSVs, and ready-to-send drafts.

Who uses it and how

• People wiring Actionbook into an MCP-compatible client — Claude (Settings → Connectors → Add custom connector), ChatGPT (Settings → Connectors), Perplexity, Hermes, OpenClaw, or any other MCP-compatible app — by pasting the hosted endpoint.
• Analysts doing multi-site sweeps, such as checking 30 company websites in parallel or collecting taglines from a 192-site portfolio set.
• Workers pulling gated numbers in one instruction, for example opening stripe.com, linear.app, and vercel.com and grabbing the pricing from each.
• Teams that turn a working task into a Recipe and rerun it on new inputs instead of re-prompting each time.
• Anyone whose data sits behind an existing login, since the extension reuses the sessions already present in their Chrome profile.

Getting started

Install the Actionbook Chrome extension from the Chrome Web Store, add https://edge.actionbook.app/mcp as a connector in your AI app, and finish setup from the dashboard at actionbook.app/dashboard, which walks you through connecting the client and confirming the extension is linked.

How it compares

No list of paid products it replaces is given in the available facts, and no directly comparable tools are named either — among similar browser-automation offerings it stands alone in this registry.

When to use it — and when not

Be aware that Actionbook depends on a hosted MCP endpoint at edge.actionbook.app and a Chrome browser running the extension, so there is no self-hosted deployment described here and the standalone @actionbookdev/cli package is deprecated; the original Actionbook is also labelled Legacy in the README. It is a poor fit if you need server-side scraping without a real browser or your own logged-in Chrome session. The repository currently carries 12 open issues, so some rough edges are still in play.

https://edge.actionbook.app/mcp

Open stripe.com, linear.app, and vercel.com, then grab the pricing from each.

Use Actionbook to understand and operate the web page.