CloudBrowser

ai.cloudbrowser由 cloudbrowser.ai 发布官方发布者v0.1.15更新于 Oct 29, 2026

Give AI agents a real cloud Chrome: open sessions, navigate, read pages, click, type and screenshot.

已验证Streamable HTTP可网页运行网页搜索与抓取浏览器自动化

概览

AI 生成的概览

托管型 MCP 服务器，为助手提供真实的云端 Chrome 浏览器，可打开会话、导航、读取页面、点击、输入和截图。

可以做什么
  • 在 CloudBrowser 账户中打开、列出、关闭并连接云端 Chrome 浏览器，启动或停止远程桌面观看
  • 导航到网址、读取页面 HTML、按选择器查找元素、点击、输入文本并截图
  • 保存浏览器会话并在之后恢复，或删除已保存的会话
  • 在对话旁的工作区中展示活动浏览器和已保存会话
  • 检测验证码，并将登录或验证环节交还给用户
适用场景
当助手需要在真实浏览器中实际操作网站时适用，例如填写表单、读取无法直接抓取的页面，或走完需要登录的流程，并且你希望实时观看或随时接管。它面向用户拥有或获授权的网站与账户，而非批量爬取。
运行要求
远程 Streamable HTTP 端点，无需本地安装或软件包。需要 CloudBrowser 账户，通过 OAuth 在客户端登录并授权，或改用 CloudBrowser API 令牌以 Authorization Bearer 请求头发送。浏览器用量按 CloudBrowser 套餐计费。
安装前请注意
OAuth 会向客户端授予浏览器控制权限；若改用 API 令牌，应保存在客户端的密钥设置中，切勿写入提示词。浏览器按套餐计量计费，恢复会话可能开启一个计费浏览器。助手可在实时会话中点击、输入和导航，会改变网站真实状态；登录、多因素验证和验证码交由用户处理，页面内容来自第三方网站。

安装

在 SourceWeft 中

  1. 打开 控制台中的 CloudBrowser，将其添加到工作区。
  2. 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/ai.cloudbrowser%2Fmcp/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。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。

{
  "mcpServers": {
    "cloudbrowser": {
      "type": "http",
      "url": "https://mcp.cloudbrowser.ai/"
    }
  }
}

README

CloudBrowser MCP Server

Give ChatGPT, Claude and your AI agents a real browser. CloudBrowser runs Chrome in the cloud. Your assistant opens a browser, navigates, reads pages, clicks, types and takes screenshots, and you can watch live or take over at any time.

  • Endpoint: https://mcp.cloudbrowser.ai (Streamable HTTP)
  • Auth: OAuth, sign in with your CloudBrowser account
  • Website: https://cloudbrowser.ai/mcp
  • Official MCP Registry: ai.cloudbrowser/mcp

What you can do

  • Open cloud Chrome browsers in your CloudBrowser account and connect to them.
  • Navigate, read page content, find elements, click, type, run scripts and take screenshots.
  • Save browser sessions and resume them later.
  • Start a remote desktop to watch the browser live or take over.
  • Detect CAPTCHAs. Login and verification walls stay with you.

Connect

The server uses OAuth. Your client opens a browser window, you sign in with your CloudBrowser account and approve access. You do not copy an API key.

Claude (claude.ai and Claude Desktop)

Open Settings → Connectors → Add custom connector, enter https://mcp.cloudbrowser.ai and select Connect.

Claude Code

bash
claude mcp add --transport http cloudbrowser https://mcp.cloudbrowser.ai

Then run /mcp in Claude Code and authenticate cloudbrowser.

Codex CLI

bash
codex mcp add cloudbrowser --url https://mcp.cloudbrowser.aicodex mcp login cloudbrowser

Cursor

Add this to ~/.cursor/mcp.json:

json
{  "mcpServers": {    "cloudbrowser": { "url": "https://mcp.cloudbrowser.ai" }  }}

VS Code

bash
code --add-mcp '{"name":"cloudbrowser","type":"http","url":"https://mcp.cloudbrowser.ai"}'

Gemini CLI

bash
gemini mcp add --transport http cloudbrowser https://mcp.cloudbrowser.ai

ChatGPT and other clients

Add a custom connector (in ChatGPT, developer mode) with the URL https://mcp.cloudbrowser.ai. Any client that supports Streamable HTTP and MCP OAuth works.

Tools

The server advertises 20 tools. Your client gets the current schemas from tools/list.

  • Workspace: browser_workspace, resume_saved_session
  • Browser lifecycle: open_browser, close_browser, get_browsers, start_remote_desktop, stop_remote_desktop
  • Saved sessions: get_saved_sessions, remove_saved_session
  • Connections: connect_to_browser, disconnect_browser, get_connected_browsers
  • Page work: navigate_to_url, get_page_content, click_element, type_text, get_elements, take_screenshot, evaluate_script, detect_captcha

Authentication and account

OAuth 2.1 with PKCE and dynamic client registration. Sign in with your CloudBrowser account. Scope: browser:control (openid email optional). Developers can send their own account token instead: Authorization: Bearer <your CloudBrowser API token>. Keep the token in your client's secret settings, never in a prompt.

You need a CloudBrowser account. Browser usage follows your CloudBrowser plan.

Links

About this repository

This repository is the public home of the hosted CloudBrowser MCP server. It holds this setup guide and the registry metadata (server.json for the official MCP Registry, glama.json for Glama). The server runs as a hosted service at https://mcp.cloudbrowser.ai. Its source code is not in this repository.

Found a problem or have a question? Open an issue.

License

The contents of this repository are available under the MIT License.

来源：README.md，提交 7bde008

工具

18

browser_workspace

Browser sessions

Show your active browsers and saved sessions in a workspace beside the conversation. Does not allocate or connect a browser. Optionally show a connected browser and its current screenshot. Use the sessionId and address from the selected browser to continue work.

输入参数browserAddress

resume_saved_session

Resume a saved browser session

Resume an existing saved session owned by the connected account. Reuses an active browser with the label, otherwise opens one metered browser with a 180-second idle timeout, restoring cookies and storage, not tabs. Only call after the user requests resume. Never retry an ambiguous allocation or change requestId to retry. Close the returned browser when finished.

输入参数label必填requestId必填

open_browser

Open one remote browser in the cloud with the connected CloudBrowser account. Use it only for websites and accounts that the user owns or is authorized to use, and follow their terms. Browsers open without stealth fingerprints.

输入参数labelbrowserheadlesskeepOpensaveSessionrecoverSession

close_browser

Close a specific browser

输入参数address必填

get_browsers

Get list of active browsers

无需输入参数。

start_remote_desktop

Start remote desktop access to a browser

输入参数address必填

stop_remote_desktop

Stop remote desktop access to a browser

输入参数address必填

get_saved_sessions

Get list of saved sessions

无需输入参数。

remove_saved_session

Delete a saved session

输入参数label必填

connect_to_browser

Connect to CloudBrowser browser for direct control

输入参数sessionId必填browserAddress必填

navigate_to_url

Navigate the connected browser to a URL that the user asked for. Use it only for websites and accounts that the user owns or is authorized to use, and follow their terms. Stop at logins, MFA and CAPTCHAs and hand control to the user; this tool does not solve CAPTCHAs.

输入参数url必填sessionId必填waitUntilbrowserAddress

get_page_content

Get the HTML content of the current page to answer the user's request about it. Use it only for websites and accounts that the user owns or is authorized to use, and follow their terms. Not for crawling or bulk data extraction from third-party websites.

输入参数sessionId必填

click_element

Click on a page element

输入参数selector必填sessionId必填

type_text

Type text into an element

输入参数text必填clearselector必填sessionId必填

get_elements

Get information about elements matching a selector

输入参数selector必填sessionId必填

take_screenshot

Take screenshot of the current page

输入参数typefullPagesessionId必填

disconnect_browser

Disconnect this control session from a browser. The browser keeps running; use close_browser to stop it.

输入参数sessionId必填

get_connected_browsers

List the browsers that this caller has connected for direct control

无需输入参数。

版本历史

1
  1. v0.1.15最新Oct 29, 2026

评分

已满足适用检查项的 100%

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
  • 已列出工具必需15 分
  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分