Browser Use
Browser agents and stealth cloud browsers for AI: run web tasks, fill forms, extract data.
作者提供的说明(英文原文)
Your own signed-in Mac browser for any agent: open, read and act in tabs. Local and free.
这三项永远不会合并成一个分数:意见、测量和人气是不同的东西。
我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。
选择你使用的 AI 应用,然后复制设置。
你需要
你用的是哪个应用?
在上方选择你的应用。我们会记住它,下次按钮会显示“安装到 …”。
打开 Claude Desktop → Settings → Developer → Edit Config,把下面的内容粘贴到 claude_desktop_config.json
{
"mcpServers": {
"yab": {
"command": "npx",
"args": [
"-y",
"yab-mcp"
]
}
}
}在终端中运行这条命令
claude mcp add yab -- npx -y yab-mcp粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"yab": {
"command": "npx",
"args": [
"-y",
"yab-mcp"
]
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
粘贴到项目中的 .vscode/mcp.json
{
"servers": {
"yab": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"yab-mcp"
]
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
这个服务器在你的电脑上运行,而 ChatGPT 只能连接互联网上的服务器。请选择其他应用。
你的选择通过 Cookie 保存在这台设备上,无需账户。
这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。
它在你的电脑上运行;安装后,工具列表会显示在你的 AI 应用中。
我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月6日
注册名称已通过域名 yetanotherbrowser.com 验证(注册中心会核实域名所有权)
我们还没有从 GitHub 读取这个仓库
我们还没有从 GitHub 读取这个仓库
它在你的电脑上运行,没有互联网端点
在你的电脑上运行,改用本地设置
它在你的电脑上运行;我们只能读取互联网端点的工具列表
与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。
它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。
注册中心没有列出必需的密钥,不过有些服务器会在首次使用时要求你登录。
它在你的电脑上运行。你的 AI 应用会在需要时启动它,数据不会经过作者的服务器(除非它自己调用在线服务)。
GitHub 仓库最后更新于–,最新版本(0.2.1)于25天前进入注册中心。
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 Node.js。
同一分类中有:Browser Use, Chrome DevTools MCP, MCP Server Browser
同一分类中其他作者的服务器,按星标数排序。
如何看卡片:★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i
Browser agents and stealth cloud browsers for AI: run web tasks, fill forms, extract data.
MCP server for Chrome DevTools
MCP server for browser use access
按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。
The MCP server of Yab, Yet Another Browser, for Claude Code, Codex, Cursor and any MCP client.
Yab, Yet Another Browser, is a free WebKit browser for the Mac. With this server, any agent gets one real browser on your Mac: the one you're signed in to. Your agent opens tabs of its own beside yours, reads pages as text, clicks, types and fills forms with your sign-ins, your cookies and your IP. There is no cloud browser, no debug port and no Chrome to set up, and it's free. A press on Send, Pay, Delete and the like waits for your yes on a card in Yab, unless you chose Bypass in Yab's settings: the browser holds that yes, not the model. Saved passwords are filled by Yab and never shown to the agent. Agent tabs live in a window off screen and come in front of you only when you ask. With Yab closed, the first call opens it hidden, with no window and no Dock tile, and it quits after half an hour with nothing to do. yab serve keeps it running that way.
Download Yab · For developers · Manual
On the assistant's own computer, run npx yab-mcp login --client dots. When Yab's Connect flow supplies a nonce, add --nonce with that value. It prints the code and returns immediately; a background process saves the connection after the person allows it in Yab. Yab on the Mac shows the eight-character code under Settings > AI > Assistants. After the person allows it there, choose who does the work:
npx and args ["-y", "yab-mcp", "--remote"]. The assistant calls open, snapshot, click, type, etc. itself, using its own model/account. --remote bridges JSON-RPC over stdio; it does not accept a natural-language prompt or start a Yab model run. Calling the task MCP tool still delegates.npx yab-mcp task "Find a table for Friday" starts a separate agent using the model selected for Tasks in Yab on the Mac. ChatGPT (via Codex) consumes the user's ChatGPT limits; Claude uses their Claude plan instead. Choose this when you intend to use that plan, including from a shell-only bot. It prints its session ID and returns; run npx yab-mcp status SESSION for the next update, or npx yab-mcp status for the newest task after a Mac reconnects. Repeat until it says done.Both modes keep the assistant's existing account access and approval settings. The connection token stays in ~/.config/yab-mcp/auth.json (mode 0600), or the path in YAB_MCP_AUTH_FILE. When the home folder is read-only, as on some bots' computers, it goes to the first writable of /workspace/shared/.yab-mcp/, /workspace/.yab-mcp/ and .yab-mcp/ in the current folder. If the computer ends the background process that waits for the person's Allow, the next task, status or --remote finishes connecting. YAB_MCP_URL selects a test server; production uses https://yab-sync.21st.workers.dev. Plain npx yab-mcp starts the local app's MCP server.
npx. A client can also run Yab directly, without Node: see Without Node.Chrome is your main browser? Yab can still be the one your agents use, usually closed. yab import chrome brings Chrome's sign-ins into Yab after you say yes in Yab, so your agent hits fewer sign-in pages.
Already use Yab? Settings › Developers adds it to Claude Code and Codex with one click, and Copy Config gives the JSON for any other client. You don't need this package for that.
The plugin brings the tools and Yab's skill:
/plugin marketplace add yetanotherbrowser/yab-mcp
/plugin install yab@yab-mcp
From a shell: claude plugin marketplace add yetanotherbrowser/yab-mcp, then claude plugin install yab@yab-mcp.
Only the tools:
claude mcp add --scope user yab-browser -- npx -y yab-mcp
If Yab's Settings › Developers already added Yab to Claude Code, skip this: you'd have the same tools twice.
The plugin, tools and skill:
codex plugin marketplace add yetanotherbrowser/yab-mcp
codex plugin add yab@yab-mcp
Or only the tools, in ~/.codex/config.toml:
[mcp_servers.yab-browser]
command = "npx" args = ["-y", "yab-mcp"] tool_timeout_sec = 900 default_tools_approval_mode = "approve"
tool_timeout_sec lets a call wait up to 15 minutes while Yab asks you something; the plugin keeps Codex's default of 5. default_tools_approval_mode stops Codex asking before every Yab tool; Yab still asks you before Send, Pay and Delete. Leave it out if you want Codex to ask too.
Or add it to ~/.cursor/mcp.json:
{
"mcpServers": { "yab-browser": { "command": "npx", "args": ["-y", "yab-mcp"] } } }
code --add-mcp '{"name":"yab-browser","command":"npx","args":["-y","yab-mcp"]}'
Or in .vscode/mcp.json:
{
"servers": { "yab-browser": { "type": "stdio", "command": "npx", "args": ["-y", "yab-mcp"] } } }
In ~/.config/devin/mcp_config.json (older Windsurf builds: ~/.codeium/windsurf/mcp_config.json):
{
"mcpServers": { "yab-browser": { "command": "npx", "args": ["-y", "yab-mcp"] } } }
Download yab-mcp- .mcpb from Releases and open it: Claude Desktop installs it as an extension. Or add this to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": { "yab-browser": { "command": "npx", "args": ["-y", "yab-mcp"] } } }
The same mcpServers entry: command npx, arguments -y yab-mcp, over stdio.
Point the client at the app itself. This is what Yab's Copy Config gives:
{
"mcpServers": { "yab-browser": { "command": "/Applications/yab.app/Contents/MacOS/yab", "args": ["mcp"], "env": { "YAB_CLI": "1" } } } }
YAB_APP if set, then the app the yab command runs (~/.local/bin/yab), then yab.app in /Applications and ~/Applications, then Spotlight (mdfind, bundle id dev.serafim.fast).yab.app/Contents/MacOS/yab mcp with YAB_CLI=1. Your client talks to Yab directly: stdin and stdout go straight through, signals are passed on, and the exit code comes back.It has no dependencies and makes no network requests of its own. npx yab-mcp --help says where it found Yab.
These come from a real tools/list of Yab, made by scripts/snapshot.mjs. The app you run is the final word: newer versions may list more.
Yab 0.0.18 lists 69 tools. Some appear only when you turn on what they need, as noted.
| Tool | What it does |
|---|---|
devtools_comments | Read user-authored comments on elements of an already inspected, non-private tab. |
devtools_console | Read recorded Console messages from an already inspected, non-private tab. |
devtools_network | Read sanitized request metadata recorded by the tab's Developer Tools. |
devtools_trace | Read the WebKit Performance recording's event names, durations and nesting. |
devtools_styles | Read the currently selected element's matched and computed styles from Developer Tools. |
native_context | Read Yab's own current settings, spaces, tabs, mail accounts, permissions and this task's native operations. |
native_actions | List native Yab actions with their strict argument schemas, targetKind, undo support and permission requirements. |
native_apply | Perform a native Yab action explicitly requested by the user: settings, spaces, tabs, page fonts or connecting Gmail. |
native_operation | Read this task's native operation by operationId, or wait for a change up to seconds (0 to 10). |
native_undo | Undo a native change of this task by its changeId (or the last undoable change). |
tabs | List the tabs open in Yab, the user's web browser: the user's and the agents'. |
open | Open a web page (URL) in a new agent tab in Yab, the user's browser, signed in to their sites, beside the user's tabs. |
spaces | List the user's spaces: sets of tabs, some signed in as someone else, with cookies of their own. |
navigate | Send an agent tab to a URL, or back, forward or reload. |
snapshot | Read a web page in a browser tab as an accessibility tree with ref ids. |
click | Click an element by ref, as a real mouse click. |
type | Replace the text in a field by ref (inputs, text areas, editors), or without ref, type into whatever has focus. |
choose | Pick an option in a by its value or label. |
press | Press keys on the focused element, as real key events. |
hover | Rest the pointer on an element, as a mouse would: menus that open on hover open. |
drag | Drag an element with the mouse: onto another element (to, its ref; the dragged one lands before it) or by dx, dy pixels. |
dialog | Answer what the page asked with alert, confirm or prompt: accept (OK) or not (Cancel), and text for a prompt. |
scroll | Scroll the page, or the box around an element by ref (lists marked scrolls), by dy pixels (negative is up). |
run | Run JavaScript that drives tabs with the same tools, several steps in one call. |
text | A tab's whole visible text, for reading articles and results. |
screenshot | A picture of a tab as it is now, for when the snapshot is not enough. |
wait | Wait for a tab to finish loading, or until its text contains for. |
eval | Run JavaScript in an agent tab (isolated from the page's own scripts, same DOM) and return its value. |
close | Close an agent tab. |
show | Put a tab in front of the user as one of theirs. |
memory_search | Search what you know about the user from earlier tasks, who they are, their accounts and sites, people, projects. |
memory_read | Read one memory page whole: USER (who the user is), MEMORY (how to work for them), TAXONOMY. |
profile | The user's profile for forms: name, headline, company, role, location, linkedin, x, github, website, telegram. |
form_answers | Answers the user typed into web forms, kept with nothing asked once the form went through. |
remember | Keep something for later tasks. |
history | The user's browsing history. |
watch | Watch a thread for someone's reply: every two hours for days. |
suggest | Hand the user what needs them, as quiet rows under a new tab's field. |
boosts | List saved site boosts, their sites, permissions and state. |
boost_read | Read the files of a saved boost or this task's trial before changing it. |
skills | Yab's skills: how particular sites work for an agent (Gmail, GitHub, LinkedIn, Amazon, …), a line each. |
skill | Read one skill whole, by name. |
handoff | Ask the user to take over a tab, sign in, pass a captcha, enter a code, and wait until they say they are done. |
ask | Ask the user to pick: a question with up to four options (they may also write their own), shown at the foot of a tab. |
| sign_in | Sign in on this tab's page with a pa
只显示了开头部分;其余内容请在仓库中阅读