Actionbook MCP
A remote MCP server and Chrome extension: the agent reads data behind your logins and operates the sites you already use
- 1,621 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The agent clicks, types and runs JavaScript in your signed-in sessions, and actions taken on sites in your name cannot be undone
- The extension reads and changes cookies and asks for access to all addresses, so data from gated pages reaches the agent
- Page content and commands pass through the hosted server edge.actionbook.app, and an outside service, Clerk, handles sign-in
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/actionbookDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent clicks, types and runs JavaScript in your signed-in sessions, and actions taken on sites in your name cannot be undone
- The extension reads and changes cookies and asks for access to all addresses, so data from gated pages reaches the agent
- Page content and commands pass through the hosted server edge.actionbook.app, and an outside service, Clerk, handles sign-in
This is third-party code. Review the repository files before installing.
What it does
Actionbook connects an AI agent to your real Chrome browser. The agent gets a single tool, `actionbook`, with subcommands of two kinds: lookup of action guides for a site (`search`, `manual`) and browser control (`goto`, `snapshot`, `click`, `type` and others). The agent learns the full command set from the tool description after connecting. The chain has three parts. The Chrome extension from the extension store keeps an outbound connection to `edge.actionbook.app`. In the agent's app you add a remote connector at `https://edge.actionbook.app/mcp`. Setup is finished in the actionbook.app dashboard, which shows that the extension is linked. Sign-in goes through the Clerk service with OAuth, and agent commands are accepted only with your account's token. This way the agent opens tabs, reads page snapshots, clicks and fills forms inside your signed-in sessions, including in parallel across several tabs. Tasks end with a finished result, for example a sourced report, a CSV table or draft emails. When a task works the way you want, you save it as a recipe and run it again with new inputs. The author has declared the earlier command line package `@actionbookdev/cli` deprecated.
Who it is for. For analysts, marketers and founders who need to collect data from dashboards and services behind a login and hand routine browser actions to an agent.
Good fit when
- You need to gather prices from several sites at once, for example open three pricing pages and put them in a table
- The data sits in a service's personal dashboard, there is no API and exporting by hand takes long
- The same browser action repeats every week and is worth saving as a recipe
- The agent must fill a form or send a draft on your behalf in an already open session
- You need a sourced report on a topic you have to read on gated pages
Not a fit when
- You need a browser that runs entirely on your computer with no outside servers: commands go through edge.actionbook.app
- Chrome has no extension installed or you cannot sign in to an actionbook.app account
- You need Firefox or Safari: the extension is built for Chrome
- You need to debug your own site with network and traces: the official Chrome DevTools MCP exists for that
Example request
Open stripe.com, linear.app and vercel.com, take the plan prices from each and build a comparison table. Use ActionbookMCP
- Transport
- http
- Authentication
- OAuth
Limitations
Part of the work goes through a hosted service: the connector attaches to `https://edge.actionbook.app/mcp`, and Clerk handles sign-in. By the extension's policy, commands are processed in server memory and not kept after the reply, but you cannot check this against server code, because the repository does not contain it.
The extension asks for the `debugger`, `tabs` and `cookies` permissions and access to all addresses. The extension's default local mode works only with the command line package, which is declared deprecated. Cloud mode is switched on by hand in the extension popup.
Commands fall into three levels. Reading runs at once. Clicks, typing and JavaScript execution also run at once, but on banking, payment, government and healthcare sites they need confirmation. Setting and deleting cookies always waits for your confirmation in the extension popup: you have 30 seconds to allow or deny, or the command is rejected.
The `skills` folder holds three skills for the earlier version: actionbook, active-research and extract. They work through the command line package. The `ACTIONBOOK_API_KEY` key is optional for them, and without it general rate limits apply.
How to disable. Remove the Actionbook connector in the agent app's settings and disable or remove the extension on Chrome's extensions page. You can sign out and revoke agent sessions in the actionbook.app dashboard.
FAQ
What do I install to get started?
The Actionbook extension for Chrome and a connector at https://edge.actionbook.app/mcp in the agent's app: Claude, ChatGPT, Perplexity, Hermes, OpenClaw or another MCP-capable one. The actionbook.app dashboard confirms the setup.
Do I need the @actionbookdev/cli package?
No. The author has declared the separate command line package deprecated and recommends the pair of a remote connector and the extension. Older guides in the docs folder still describe installation through `npx -y @actionbookdev/mcp@latest`.
How does the extension protect sensitive sites?
The allowed Chrome debugging commands are restricted to a list. For banking, payment, government and healthcare addresses, clicks and typing are raised to the level that needs confirmation. The confirmation comes in the extension popup and lasts 30 seconds.
Does the extension see my password?
No. Sign-in goes through a hosted Clerk page, and the extension receives an access token and a refresh token from it and keeps them in `chrome.storage.local`. The access token lives from a few minutes to an hour.
What is a recipe?
A saved task that the agent in the Chrome side panel completed the way you wanted. You run it again with different inputs.
What license is the project under?
Apache-2.0. The repository holds the source code of the extension, the command line version and the packages for the JS SDK, MCP and the Dify and OpenClaw plugins. The code of the hosted server is not in the repository.
Discussion
No questions or reviews yet.
Related
MCP serversA CLI and MCP server that lets an agent drive your real Chrome with its logged-in sessions: 103 commands across 36 platforms plus browser automation
A Chrome extension with an MCP server: the agent drives your everyday browser with its sessions, tabs, bookmarks and history
An MCP and CLI that let an agent control your real Chrome, with its own logins and extensions, through Playwright
An MCP bridge to control your browser from a coding agent: a Chrome extension and a set of tab and page automation tools
A local browser automation platform: MCP server, CLI and skills so an agent can drive Chrome, fill forms and pull data from pages
An MCP server that drives your already open Chrome and stops to ask you about a 2FA code, a CAPTCHA, or a choice only you can make
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Actionbook (this entry) | yes | OAuth | 1,621 | |
| Browser MCP | yes | not required | 7,246 | an MCP server and Chrome extension to drive an already open browser, all running locally |
| Chrome DevTools MCP | yes | not required | 54,368 | the Chrome DevTools team's official MCP server: network, console and performance traces of a live Chrome |
README in short
The project description opens with words on the new Actionbook in the Chrome side panel and an install button for the extension store. A section on the earlier version explains how working from site guides differs from ordinary page parsing: slow, brittle and one page at a time, against fast, reliable and parallel. Then come three install steps, a quick start with a sample request, a description of the single tool and links to documentation, community and a form to request a new site. The repository has documentation with several guides, console and SDK packages, plugins and three skills.