timesfm
TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.
Drive a real browser with shell-like text commands (cd/ls/cat/click/type), using far fewer tokens.
Skills that teach the AI this kind of work; they work alongside this one
TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.
Open-source intelligence for the global theater. Track everything from the corporate/private jets of the wealthy, and spy satellites, to seismic events in one unified interface. Hook an AI agent up to have it parse through data and find previously unseen correlations. The knowledge is available to…
scrape data from Google Maps. Extracts data such as the name, address, phone number, website URL, rating, reviews number, latitude and longitude, reviews,email and more for each place
Lists are ordered by a usage score (out of 100) built from these parts
| Popularity | 0/80 | GitHub stars or weekly downloads: the larger counts 60, the smaller 20 |
|---|---|---|
| Freshness | 12/12 | Full marks for a release or code change within 30 days |
| Still answering | 0/8 | The cloud endpoint answered at the last check |
| Usage score | 12/100 |
Quality: Passed 3/3 basic checks (shown, not used for ordering) · Momentum: No data yet (Used only on the Trending tab)
Pick the AI app you use, then copy the settings.
Let your AI install it with one sentence
Please install WebShell. Follow https://mcp.popnx.com/i/felixlin789-webshell.md, show me the commands and wait for my OK before running them.Paste it into an AI that can work on your computer (such as Claude Code or Cursor). It reads our steps, shows you the commands and waits for your OK. No key, no sign-up. · See the full text your AI reads
Passed 3/3 checks · Runs on your computer · ✓ Commands checked Nov 18, 2026
First: which AI do you use?
Pick your app and the steps appear below.
Works with Claude, with some setup first
Install Node.js on your computer first (one time) · Only the Claude desktop app (Windows or Mac), not the web or phone
Copy this message into a Claude chat. The AI checks your computer, tells you which file it will change, and waits for your OK every time.
Please install the MCP server WebShell for Claude on this computer, in this order: 1) Check Node.js with node -v. If it is missing, tell me to install it from https://nodejs.org/en/download; do not install it yourself. 2) Use only this config (already checked by PopNx MCP): { "mcpServers": { "webshell": { "command": "npx", "args": [ "-y", "webshell-mcp", "mcp" ] } } } 3) Tell me the full path of the file you will change, or the command you will run, and wait for my OK. 4) No key is needed. 5) Do what we agreed, then check that the WebShell tools really show up.This message has no key in it. If a key is needed, the AI will ask you to paste it yourself into the file or the app’s field.
Already installed? Check first · Open Terminal, type this command and press Enter
node -vMac: press ⌘ + Space, type Terminal, press Enter
Windows: open Start, type Windows PowerShell, then Open
Not installed: Open the official download page, pick the LTS version and the click-through installer: .pkg on Mac · .msi on Windows. Double-click and follow it to the end. No curl command needed
Open the Node.js download page ↗
When it’s done, close Terminal, open it again and run the same command. You should see a version number: node -v
Install once; it works for everything that needs Node.js
SettingsDeveloperEdit Config
{
"mcpServers": {
"webshell": {
"command": "npx",
"args": [
"-y",
"webshell-mcp",
"mcp"
]
}
}
}Use WebShell and tell me what you can do for me, then show me one thing✓ If Claude says it is using WebShell or shows a WebShell tool in the answer, you are connected
Menu names checked against Anthropic’s help page on Nov 11, 2026 · your screen may differ slightly by app version
ChatGPT cannot connect WebShell yet
WebShell runs on your computer, but ChatGPT can only connect servers that run on the web
A · Use an app that can connect WebShell
B · Others in Web & browsers that work with ChatGPT
Picked from the most used in the same category, only ones that connect by address with no key
Works with Claude Code, with some setup first
Install Node.js on your computer first (one time)
Copy this message into a Claude Code chat. The AI checks your computer, tells you which file it will change, and waits for your OK every time.
Please install the MCP server WebShell for Claude Code on this computer, in this order: 1) Check Node.js with node -v. If it is missing, tell me to install it from https://nodejs.org/en/download; do not install it yourself. 2) Use only this config (already checked by PopNx MCP): claude mcp add --scope user webshell -- npx -y webshell-mcp mcp 3) Tell me the full path of the file you will change, or the command you will run, and wait for my OK. 4) No key is needed. 5) Do what we agreed, then check that the WebShell tools really show up.This message has no key in it. If a key is needed, the AI will ask you to paste it yourself into the file or the app’s field.
Already installed? Check first · Open Terminal, type this command and press Enter
node -vMac: press ⌘ + Space, type Terminal, press Enter
Windows: open Start, type Windows PowerShell, then Open
Not installed: Open the official download page, pick the LTS version and the click-through installer: .pkg on Mac · .msi on Windows. Double-click and follow it to the end. No curl command needed
Open the Node.js download page ↗
When it’s done, close Terminal, open it again and run the same command. You should see a version number: node -v
Install once; it works for everything that needs Node.js
Mac: press Command+Space and type Terminal · Windows: open PowerShell
claude mcp add --scope user webshell -- npx -y webshell-mcp mcpYou should see WebShell listed as connected
Use WebShell and tell me what you can do for me, then show me one thing✓ If Claude Code says it is using WebShell or shows a WebShell tool in the answer, you are connected
Menu names checked against Anthropic’s help page on Nov 11, 2026 · your screen may differ slightly by app version
Works with Cursor, with some setup first
Install Node.js on your computer first (one time)
Copy this message into a Cursor chat. The AI checks your computer, tells you which file it will change, and waits for your OK every time.
Please install the MCP server WebShell for Cursor on this computer, in this order: 1) Check Node.js with node -v. If it is missing, tell me to install it from https://nodejs.org/en/download; do not install it yourself. 2) Use only this config (already checked by PopNx MCP): { "mcpServers": { "webshell": { "command": "npx", "args": [ "-y", "webshell-mcp", "mcp" ] } } } 3) Tell me the full path of the file you will change, or the command you will run, and wait for my OK. 4) No key is needed. 5) Do what we agreed, then check that the WebShell tools really show up.This message has no key in it. If a key is needed, the AI will ask you to paste it yourself into the file or the app’s field.
Already installed? Check first · Open Terminal, type this command and press Enter
node -vMac: press ⌘ + Space, type Terminal, press Enter
Windows: open Start, type Windows PowerShell, then Open
Not installed: Open the official download page, pick the LTS version and the click-through installer: .pkg on Mac · .msi on Windows. Double-click and follow it to the end. No curl command needed
Open the Node.js download page ↗
When it’s done, close Terminal, open it again and run the same command. You should see a version number: node -v
Install once; it works for everything that needs Node.js
Use WebShell and tell me what you can do for me, then show me one thing✓ If Cursor says it is using WebShell or shows a WebShell tool in the answer, you are connected
Menu names checked against Cursor’s help page on Nov 11, 2026 · your screen may differ slightly by app version
Works with VS Code, with some setup first
Install Node.js on your computer first (one time)
Copy this message into a VS Code chat. The AI checks your computer, tells you which file it will change, and waits for your OK every time.
Please install the MCP server WebShell for VS Code on this computer, in this order: 1) Check Node.js with node -v. If it is missing, tell me to install it from https://nodejs.org/en/download; do not install it yourself. 2) Use only this config (already checked by PopNx MCP): { "servers": { "webshell": { "type": "stdio", "command": "npx", "args": [ "-y", "webshell-mcp", "mcp" ] } } } 3) Tell me the full path of the file you will change, or the command you will run, and wait for my OK. 4) No key is needed. 5) Do what we agreed, then check that the WebShell tools really show up.This message has no key in it. If a key is needed, the AI will ask you to paste it yourself into the file or the app’s field.
Already installed? Check first · Open Terminal, type this command and press Enter
node -vMac: press ⌘ + Space, type Terminal, press Enter
Windows: open Start, type Windows PowerShell, then Open
Not installed: Open the official download page, pick the LTS version and the click-through installer: .pkg on Mac · .msi on Windows. Double-click and follow it to the end. No curl command needed
Open the Node.js download page ↗
When it’s done, close Terminal, open it again and run the same command. You should see a version number: node -v
Install once; it works for everything that needs Node.js
Use WebShell and tell me what you can do for me, then show me one thing✓ If VS Code says it is using WebShell or shows a WebShell tool in the answer, you are connected
Menu names checked against VS Code’s help page on Nov 11, 2026 · your screen may differ slightly by app version
Claude Code
claude mcp add --scope user webshell -- npx -y webshell-mcp mcpClaude Desktop (claude_desktop_config.json)
{
"mcpServers": {
"webshell": {
"command": "npx",
"args": [
"-y",
"webshell-mcp",
"mcp"
]
}
}
}Cursor (~/.cursor/mcp.json)
{
"mcpServers": {
"webshell": {
"command": "npx",
"args": [
"-y",
"webshell-mcp",
"mcp"
]
}
}
}VS Code (mcp.json)
{
"servers": {
"webshell": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"webshell-mcp",
"mcp"
]
}
}
}These settings are generated from the maker’s registry entry. If the README says otherwise, follow the README.
Based on how this server connects and on each app’s own docs, checked Nov 10, 2026.
Run on your computer in
ClaudeGooseLM StudioMsty StudioCherry StudioCursorVS Code + GitHub CopilotVisual StudioWindsurf (Devin Desktop)ZedKiroClineContinueJetBrains AI AssistantClaude CodeCodex CLIGemini CLIOpenCodenodecodeWarpLibreChatRaycast
Runs on your computer; the tool list appears in your AI app after you install it.
We check information anyone can see; we have not read the code. Checks that do not apply are left out of the count. · Checked Nov 19, 2026
The registry name was verified with the GitHub account FelixLin789, which owns the repo
Code last changed yesterday
Uses MIT
This one runs on your computer; there is no internet endpoint
Runs on your computer and uses local settings instead
Runs on your computer; we can only read tool lists from internet endpoints
No entries on OSV.dev for webshell-mcp@0.3.0
Checks only this package version, not every dependency · checked Nov 18, 2026
If you made this, paste this code into your README. The badge links back to this page.
[](https://mcp.popnx.com/en/servers/felixlin789-webshell)Ask, share a tip or report that it does not work. Sorted by what people marked helpful.
No posts yet. Be the first to ask or share.
Posts here are written by users, not the PopNx team.
No reviews yet
Compared with every active server in this category in the registry. Stars count only when the publisher owns the repo.
It passed 3 of 3 applicable basic checks. We have not read the code. Before adding a key or token, see what access the server asks for and use a key with only the permissions you need.
The registry does not list a required key, though some servers ask you to sign in on first use.
It runs on your computer. Your AI app starts it when needed, and data does not pass through the maker’s servers (unless it calls an online service itself).
The GitHub repo last changed yesterday, and the latest version (0.3.0) reached the registry yesterday.
Apps that support MCP, such as Claude Desktop, Claude Code, Cursor and VS Code. You also need Node.js on your computer.
In the same category: Browser Use, changedetection.io (official), Apify MCP Server, Pomerado, Actionbook
Same category, other makers, commands already checked. Official ones first, then by stars.
Reading a card:↓ Downloads per weeki★ GitHub starsiThree pillars: reviews · checks · usageiUpdated = latest version published to the registryi
Lets AI do web tasks for you, such as filling forms, searching and extracting data, in the maker’s cloud browsers or on your computer.
Official changedetection.io MCP: web page change tracking, back-in-stock and price alerts.
Extract data from any website with thousands of scrapers, crawlers, and automations on Apify Store ⚡
Connect your agents to any website. The cheaper, faster and more reliable alternative to browser use
Let your AI agent operate any website through a real, logged-in browser session.
| This one WebShell | Browser Use | changedetection.io (official) | Apify MCP Server | Pomerado | Actionbook | |
|---|---|---|---|---|---|---|
| Runs where | Installs on your computer | Cloud or install on your computer | Runs in the cloud, nothing to install | Runs in the cloud, nothing to install | Runs in the cloud, nothing to install | Runs in the cloud, nothing to install |
| Needs a key | No | Yes | No | Yes | No | No |
| Add with + in Claude | No | Yes | No | No | No | No |
| Basic checks | 3/3 | 5/5 | 3/3 | 4/4 | 4/4 | 4/4 |
| License | MIT | MIT | Apache-2.0 | MIT | MIT | Apache-2.0 |
| Price | No cost listed in the registry | No cost listed in the registry | No cost listed in the registry | No cost listed in the registry | No cost listed in the registry | No cost listed in the registry |
Copied from the maker’s repo as written. PopNx MCP has not reviewed or endorsed it. HTML, images and scripts are removed. Links lead off this site.
Let AI agents drive a real browser with shell-like text commands, using a fraction of the tokens of snapshot- or screenshot-based tools.
cd into a site, ls what you can click or fill, cat the text, click / type by number. Output is plain text, kept short on purpose: repeated items are merged, long pages can be read section by section, and after a click or keystroke you get only the text that newly appeared, not the whole page again. It can drive your own Chrome with your existing logins, and it ships as a CLI and as an MCP server with a single tool.
$ ws cd github.com/microsoft/vscode
GitHub - microsoft/vscode: Visual Studio Code · GitHub — https://github.com/microsoft/vscode $ ws ls | grep -w Code 37 button Code $ ws click 37 GitHub - microsoft/vscode … — https://github.com/microsoft/vscode 新出现: ← "newly appeared": just the menu that opened Clone HTTPS GitHub CLI Download ZIP
| Page / action | WebShell | agent-browser snapshot -i | Full accessibility snapshot (Playwright MCP style) |
|---|---|---|---|
| GitHub repo page: list what can be clicked | ~1,200 | ~7,200 | ~10,900 |
| Bootstrap docs page: list what can be clicked | ~500 | ~2,900 | ~11,700 |
| Google Flights search page | ~600 | ~1,100 | ~2,400 |
| After clicking GitHub's "Code" menu | 63 (only the new text) | ~7,200 (snapshot again) | — |
| Wikipedia "Alan Turing": read two sections | ~2,600 (outline, then sections) | — | ~158,000 (Playwright MCP snapshot) |
Counts are estimates (about 4 characters per token for English text); a screenshot costs roughly 1,000–2,000 tokens per image and cannot see anything off-screen.
Requires Node.js 22+ and Chrome. No dependencies.
npm install -g webshell-mcp # then `ws` works anywhere
As an MCP server, for example in Claude Code:
claude mcp add webshell -- npx -y webshell-mcp mcp
Or in any MCP client's JSON config:
{ "mcpServers": { "webshell": { "command": "npx", "args": ["-y", "webshell-mcp", "mcp"] } } }
login fills them from an encrypted local password book (Windows for now) that only you can edit..ssh, browser profiles, autostart folders).The full documentation below is in Chinese. Run ws with no arguments for the command list.
让 AI 像用命令行一样操控网页:cd 进网站,ls 看能点、能填什么,cat 读内容。 默认只给文字,省 token;跑在你自己的电脑上,可以直接用你自己的 Chrome 和里面的登录状态。
$ ws cd news.ycombinator.com
Hacker News — https://news.ycombinator.com/ $ ws ls | grep -i login 10 link login $ ws cat | grep AI United Airlines bans AI agents from booking flights on behalf of travelers (cbsnews.com)
零依赖,需要 Node.js 22+ 和 Chrome(Windows、macOS、Linux;密码本目前只支持 Windows)。代码文件:
webshell.mjs:入口:命令行、交互模式、MCP 服务器(把命令传给后台进程)daemon.mjs:后台进程:一直连着浏览器,执行所有命令page.mjs:注入到网页里执行的代码(列元素、读文字、安全检查)common.mjs:两边共用的路径、版本号、浏览器连接text.mjs:目录和段落、grep / head / extract 这些文字处理vault.mjs:密码本从 npm 安装,之后在任何地方都能用 ws:
npm install -g webshell-mcp
也可以从源码装:克隆这个仓库,在文件夹里运行一次 npm link;或者不安装,把下面的 ws 换成 node webshell.mjs。
chrome://inspect/#remote-debugging,打开开关。ws 时,Chrome 会弹出“允许远程调试”的对话框,点“允许”。之后 WebShell 在后台一直连着,不会再问。只有后台进程重新启动时(重启了 Chrome、执行了 ws exit、或者更新了代码)才会再问一次。
不打开这个开关时,WebShell 会启动一个自己专用的 Chrome 窗口(配置在 ~/.webshell/profile),登录状态也会一直记住。
想用云端浏览器:把环境变量 WS_CDP 设为它的连接地址(ws:// 或 wss://)。
注意:打开开关并点了“允许”之后,AI 能访问你 Chrome 里所有已登录的网站(邮箱、网银等)。不用时执行 ws exit 断开,或者把开关关掉。
ws cd <网址|/路径|..> 打开网站;不带参数显示当前位置
ws ls [-l] [--all] [--form] 列出能点、能填的东西;-l 带链接地址,--all 按原顺序逐条列出; --form 只列表单里的框和按钮,按小标题分组 ws cat [网址...] [--field 名字] 读页面文字;给多个网址就并行读;--field 只取一个字段,如 price ws cat --outline / --section 标题 正文多时先看目录,再只读需要的那一段 ws click <编号|文字> [--accept] 用鼠标点击;点出新标签页会自动跟过去;--accept:弹出确认框时点“确定” ws expand <编号...> [--section 标题] 批量展开详情,等页面稳定一次,再读取正文或指定段落 ws expand <编号...> [--section 标题] --extract '正则' --check '标记' 展开状态和指定字段齐了就返回 ws type <框> <内容> [<框> <内容>...] [--enter] 输入(内容为 "" 则清空),可回车提交;对下拉框则是选中该选项。 一次可以填好几个框(如 type 姓名 张三 邮箱 a@b.c),只在最后等一次页面,回车按在最后一个框 ws login 登录:用你为这个网站选的方式(Google / Apple 等,或账号密码),见下面“自动登录” ws scroll [屏数] 向下滚几屏(负数向上):有些内容滚到才加载,新加载的文字直接带回来 ws tab [编号|close] 列出标签页 / 切换过去 / 关掉当前这个 ws upload <文件...> [--to 编号|文字] 上传文件;页面上有好几个上传框时用 --to 指定 ws download <编号|文字> [文件夹] 点击并下载,返回存到了哪里(默认存进“下载”文件夹) ws screenshot [文件] [--full] 截图(--full 截整页) ws pdf [文件] 把页面存成 PDF ws mobile on|off 用手机的屏幕大小和身份打开网页 ws save 名字 / ws run 名字 [参数] 把最近一次 cd 之后的操作存成一条命令 / 重放 ws exit 断开(专用窗口会被关掉;你自己的 Chrome 不会被关)
ls 和 cat 会穿过 shadow DOM 和内嵌页面(iframe,包括别的网站嵌进来的)。展开、勾选、选中、不可用这些状态直接写在 ls 的最后一列。
长表单用 ls --form:只列输入框、下拉框、勾选框和表单里的按钮,按表单里的小标题(fieldset 的标题,或前面最近的页面标题)分组,像这样:
## Shipping address
12 text First name 13 text City
20 text First name
收货地址和账单地址里都有的“First name”一眼分得清,不用先 grep。
ls 会把出现至少 5 次的相同列表项合成一行,保留每个元素的编号。例如合并行的第一列是 12,19,26 时,可用 click 19 操作编号 19 的元素。引用角标([1]、[note 2])和“Jump up”这类只在本页内跳转的回跳链接不列出,最后一行会说省略了几个;网站用 href="#..." 做的标签页、折叠按钮照样列出。需要看各元素原本的位置和上下文,或要看被省略的那些时,用 ls --all。
click、type、scroll、upload 在同一份网页里操作后,会在“新出现”下面列出新增的文字,例如展开的内容、输入候选、滚动加载的新内容。新增预览最多 2,000 字,更多内容可用 cat --outline 或 cat | grep 找。若操作打开了新页面,用 cat 读取。
点击默认像人一样用鼠标点;如果按钮被弹窗、横幅挡住了,或者在别的网站嵌进来的页面里,就改用脚本点击。
流程文件是纯文本,在 ~/.webshell/macros/名字.jsonl,一行一步,可以手改——比如把搜索词改成 $1,以后就能 ws run 名字 新的搜索词。
AI 从头到尾看不到你的密码:
ws login 时,屏幕上会弹出 Windows 自带的账号密码输入窗口,你本人填一次就行。ws login 会自己解开密码、填好并提交,不再弹窗。也可以提前存好:ws pass save login.example.edu(同样是弹窗输入)。
~/.webshell/vault.json,只有你这个 Windows 账户能解开,文件被拷走也没用。ws pass list 看存了哪些网站;ws pass rm 网站 删掉一个。login.example.edu 上,账号就存在这个名字下),弹窗里会写明是哪个网站。验证码(两步验证)还需要你本人输入:手机上的验证码 AI 拿不到。
登录页上有“Continue with Google”“使用 Apple 登录”这类按钮(认 Google、Apple、Microsoft、GitHub、Facebook)时,AI 第一次运行 ws login,屏幕上会弹出一个窗口列出这个网站能用的登录方式,你本人选一种:
你的选择会记在密码本里(ws pass list 显示为“用 Google 登录”),以后自动用同一种方式;想换就 ws pass rm 网站。
第三方登录要用你自己的 Chrome(里面已经登录了 Google):专用窗口里 Google 常会拒绝登录(“此浏览器可能不安全”)。Google 的“允许某应用访问你的账号”这类授权页、两步验证,仍然要你本人来。
能跑终端的 AI(Claude Code 等):告诉它“用 ws 命令操控网页,先运行 ws 看用法”。输出是纯文本,可以接系统的管道:ws cat | grep -i price。
MCP:WebShell 自己就是一个 MCP 服务器,只有一个工具 ws,参数是一行命令。以 Claude Code 为例:
claude mcp add webshell -- npx -y webshell-mcp mcp
MCP 里自带 | grep [-i] [-v] [-c] [-C 行数] 和 | head -N,还能用分号一次做几步:cd github.com; ls | grep -i login。没加过滤器时,每条命令最多返回 200 行。screenshot 会把图片直接交给 AI。文件路径请写完整路径。更新了 WebShell 的代码之后,要重启用它的程序(比如 Claude Desktop)。
交互模式和 MCP 都支持 | extract '正则'。比如页面上的价格标签和金额分两行时,cat | extract '^Price from\n(.+)$' | head -3 只返回前三个起价。用 \n 匹配换行,用括号选字段。每条匹配的字段按捕获顺序用 Tab 分隔。省略括号时返回匹配原文。
提取列表时,可加 --check '每条记录的标记',例如 | extract '^Price from\n(.+)$' --check '^Price from$' | head -3。校验在 head 之前运行。每个标记须落在一条匹配内,捕获的字段须有值。漏记录、缺字段或找不到标记时会报错。按任务选择每条记录都带的标记,校验覆盖的是当前读到的文本。
并行核对详情用 cat URL1 URL2 --extract '字段正则' --check '记录标记',可以加 --section '标题'。如果已知每页应有几条,再加 --expect 条数,字段和条数齐了立即返回。也可用 cat 列表URL --follow '链接文字正则' --limit 5 --extract '字段正则' --check '记录标记' 从当前列表选链接,省略列表网址则用当前标签页。链接去重并按页面顺序读取,每批最多打开 8 页。
每个结果带请求网址、跳转后的实际网址(如有)、状态、匹配数/标记数和已识别的下一页。ok 表示指定范围通过校验。约 10 秒内正文已可读但字段没齐时返回 incomplete,页面未及时就绪时返回 timeout。登录跳转、HTTP 429 和其他访问失败分别标记。列表还会给出本次选中数/当前匹配链接数。next=未知 表示没有识别到标准分页标记。可按失败来源单独补读。
批量读详情时,先用 ls 找编号,再用 expand 12 19 26 --section "Top flights" | extract '正则' --check '标记'。同一批的编号来自同一个网页或内嵌页面。expand 识别原生详情和带展开状态的按钮,已展开的会保持打开。按钮只接受真实鼠标点击或没有公开展开状态时,用 click。
每个按钮对应一条记录时,可以把提取条件直接交给 expand,例如 expand 12 19 26 --section "Top flights" --extract '正则' --check '标记'。它等到所有按钮已展开、指定范围内的标记都匹配、捕获字段都有值,且记录数等于按钮数,就返回这些字段。用能区分完整详情和占位内容的规则。超过约 10 秒仍未齐会报错。读取全文或记录数不固定时,用普通 expand。
交互模式:ws shell,写法和 MCP 里一样,适合自己手动试。
同一个页面,交给 AI 的内容有多大(实测,2026 年 11 月):
| 页面 | 整棵无障碍树(原始 JSON) | ws ls | ws cat | 先 --outline 再读需要的 --section |
|---|---|---|---|---|
| Wikipedia“Alan Turing” | 约 15.8 万 token(Playwright MCP 快照) | 约 1.25 万(合并前) | 约 2.8 万 | 约 2,600 |
| Google Flights 搜索结果 | 约 1.5 万 token(Playwright MCP 快照) | — | 约 550 | — |
真正省的地方是过滤:AI 很少需要整页,ws ls | grep -i search 通常只返回一两行。作为对比,一张截图约 1,400–2,000 个 token。
两项省 token 功能在 2026 年 11 月 13 日用专用 Chrome 和 o200k_base 实测。数字只计算返回的文字,不含命令和模型自身的消耗。
| 操作 | 完整输出 | 精简输出 | 减少 |
|---|---|---|---|
| Wikipedia“Alan Turing”元素列表 | 16,388 | 14,384 | 12.2% |
| Hacker News 元素列表 | 1,629 | 1,484 | 8.9% |
| GitHub VS Code 展开 Code 菜单 | 1,252(位置 + cat) | 63(位置 + 新出现的文字) | 95.0% |
GitHub 和 BBC 新闻的元素列表没有达到合并门槛。GitHub 的 Code 菜单位于正文区之外,新增文字还带回了普通 cat 没读到的 HTTPS 和 Download ZIP。
之后把合并门槛降到 5 次、不再列出引用角标,又测了一轮(同日,按字符数估算:英文约 4 个字符 1 个 token,所以只和同一轮的数字比):
| 操作 | 旧做法:动作 + 再 cat 整页 | 现在:动作的输出 | 减少 |
|---|---|---|---|
| Wikipedia“Alan Turing”元素列表 | 10,711(门槛 10) | 9,029(门槛 5,省略 586 个引用角标) | 16% |
| Bootstrap 文档:展开折叠面板 | 4,737 | 141 | 97% |
| GitHub VS Code:展开 Code 菜单 | 1,437 | 63 | 96% |
| Google Flights:输城市不回车 | 441 | 107(直接带回全部候选) | 76% |
| Hacker News:点 More(跳到新页面) | 975 | 12(只报位置,要看再 cat) | — |
工具说明(每次对话都要加载)从约 1,040 压到约 730 个 token。
ws login 让工具从密码本里填,看不到密码本身。click ... --accept 才点“确定”,但提到付款、下单的确认框永远不替你确认。upload、download、screenshot、pdf 不碰敏感位置:WebShell 自己的数据目录(密码本、浏览器登录资料)、点开头的配置目录(.ssh、.aws 等)、浏览器资料目录和开机启动目录,碰到就拒绝(退出码 2);截图和 PDF 也不覆盖已有文件。这是按位置名拦的,拦不住所有私密文件:别让 AI 在不可信的网站上随意上传。screenshot 看。--field 是猜的:先找网站自带的结构化数据,再找同名的 class/id,最后退到含这个词的第一行。screenshot --full 不适合特别长的页面(图片会非常大)。[Console]::OutputEncoding = [Text.Encoding]::UTF8。WS_CDP:连到别处已经开着的浏览器(比如云端浏览器),退出时不关它。WS_BROWSER=own:即使打开了开关,也用 WebShell 专用窗口。WS_CHROME:Chrome 不在默认位置时,设为它的完整路径。WS_HOME:数据目录,默认 ~/.webshell。WS_HEADLESS=1:专用窗口不显示界面(自检用)。npm test
会起一个本地小网站,用无界面的专用 Chrome 把每条命令跑一遍(包括用假账号测试自动登录),不会碰你自己的 Chrome 和密码本。