
Actionbook
app.actionbook由 actionbook.app 发布官方发布者v1.0.0更新于 Oct 26, 2026
Let your AI agent operate any website through a real, logged-in browser session.
概览
远程 MCP 连接器,让 AI 助手查询操作手册并驱动你已登录的 Chrome 操作网站,交付完成的工作成果。
- 可以做什么
- 查询网站的操作手册(search、manual),让助手按步骤执行而无需解析页面
- 控制真实的 Chrome 会话:打开页面、读取快照、点击与输入,并可同时操作多个标签页
- 在已登录状态下工作,能直接调用 API 时用 API,否则用界面自动化
- 产出可直接使用的结果,例如带来源的报告、干净的 CSV 或待发送的草稿
- 把跑通的任务保存为 Recipe,换新输入再次运行
- 适用场景
- 适合需要助手在你已登录的网站上完成多步骤工作的场景,例如从多个网站抓取价格、采集登录后才能看到的数据,或在大量页面上重复同一流程。
- 运行要求
- 在支持 MCP 的客户端中添加远程 MCP 端点作为连接器,使用 OAuth 登录,并安装 Actionbook Chrome 扩展并在仪表盘中完成关联。无需安装命令行工具;独立 CLI 包已被说明为弃用。
安装
在 SourceWeft 中
- 打开 控制台中的 Actionbook,将其添加到工作区。
- 连接 所需的 凭据。它们会被私密存储,绝不会在公开页面上展示。
- 为需要使用其工具的对话启用该服务。
Web executable,通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。
交给智能体安装
把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体,或 SourceWeft 对话中。智能体会读取该服务器的安装指南,先向你展示发布者、端点和登录方式,经你同意后再把它添加到你的客户端。密钥不会写进任何命令。
阅读 https://sourceweft.com/mcp/app.actionbook%2Factionbook/install.md,并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前,先告诉我它的发布者、端点和登录方式,等我确认。不要把 API 密钥写进命令或我会提交的文件;问我密钥应从哪里获取。其他 MCP 客户端
把它添加到你客户端的 mcpServers 配置中。 客户端会在首次连接时完成 OAuth 登录。
{
"mcpServers": {
"actionbook": {
"type": "http",
"url": "https://edge.actionbook.app/mcp"
}
}
}README
[Meet your AI Browser Coworker]
Actionbook is an AI agent in your browser. It reads data behind your logins,
operates websites you already use, and delivers finished work, not answers.
Meet the New Actionbook
Actionbook now ships as an AI agent in your Chrome side panel. It works beside you on the page you are on, reads data behind your logins, and delivers finished work, not answers: a sourced report, a clean CSV, or ready-to-send drafts. When a task works the way you want, save it as a Recipe and run it again with new inputs.
[Add to Chrome - It's Free] [Visit actionbook.app]
Actionbook (Legacy)
Actionbook turns the websites you work in every day into something your AI agent can actually operate. Direct API requests when possible, UI automation when not, with login handled. Fast and resilient.
Why Actionbook?
❌ Without Actionbook
- Slow. Agents take a snapshot after every single step, parse the page, then decide what to do next. Searching one room on Airbnb takes 5 minutes.
- Brittle. Modern websites use virtual DOMs, streaming components, and SPAs. Agents don't understand these rendering mechanisms, so they fail on dropdowns, date pickers, and login walls.
- One at a time. Your agent finishes one page before it can start the next. Need to check 30 company websites? That's 30 rounds, one after another.
✅ With Actionbook
- 10x faster. Action manuals tell agents exactly what to do. No parsing, no guessing.
- Accurate. Direct API calls when possible, UI fallback when not. Login handled either way.
- Concurrent. Stateless architecture. Operate dozens of tabs in parallel.
See an agent visits 192 First Round portfolio company websites and collects their taglines in 2 minutes. (Video is not sped up)
https://github.com/user-attachments/assets/35079a19-7236-47a8-87ed-3edf6436c2bf
Installation
Actionbook runs as a remote MCP connector paired with the Actionbook Chrome extension — no CLI install required. Your AI agent connects to the hosted MCP endpoint, and the extension drives your real Chrome so agents reuse your logged-in sessions.
👉 For the full step-by-step setup — with per-client instructions and screenshots — visit the dashboard. The three steps below are the quick overview.
1. Install the Chrome extension from the Chrome Web Store.
2. Add Actionbook as a connector in your AI app. Open your app's connector settings — Claude (Settings → Connectors → Add custom connector), ChatGPT (Settings → Connectors), Perplexity, Hermes, OpenClaw, or any other MCP-compatible app — and paste the Actionbook endpoint:
3. Finish setup from the dashboard — it walks you through connecting your client and confirms the extension is linked.
Note: The standalone
@actionbookdev/clipackage is deprecated. Use the remote connector + Chrome extension above.
Quick Start
Once the extension is installed and your client is connected (see Installation), just talk to your agent — Actionbook handles the rest:
Behind the scenes the agent uses Actionbook to fetch action manuals and operate your real Chrome — opening tabs, reading snapshots, clicking, and filling forms across pages concurrently. To nudge it explicitly, add this to your prompt:
Examples
Explore real-world examples in the Examples Documentation.
Available Tools
Actionbook exposes a single actionbook MCP tool whose subcommands cover action lookup (search, manual) and browser control (goto, snapshot, click, type, …). Your agent discovers the full command set from the tool description once connected.
Chrome Extension
Install the extension from the Chrome Web Store. Once you sign in from the dashboard, it connects to the hosted MCP endpoint automatically.
Documentation
For comprehensive guides, API references, and tutorials, visit our documentation site:
Stay tuned
We move fast. Star Actionbook on Github to support and get latest information.
Join the community:
- Chat with us on Discord - Get help, share your agents, and discuss ideas
- Follow @ActionbookHQ on X - Product updates and announcements
Contributing
- Read the Contributing Guide - See repository setup, package layout, and validation workflows for the public repo.
- Request a Website - Suggest websites you want Actionbook to index.
License
See LICENSE for the license details.
来源:README.md,提交 20ed5b5
工具
0版本历史
1- v1.0.0最新Oct 19, 2026
评分
已满足适用检查项的 85%
该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。
- 近 30 天内可连通必需25 分
- 已声明认证方式必需10 分
- 已列出工具15 分
请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。
- 描述必需15 分
- 图标10 分
- 文档15 分
- 许可证10 分