MCP服务器在本地浏览器中抓取X、LinkedIn和Hacker News的公开帖子，而该浏览器从未登录
概览
一个MCP服务器会在你自己的电脑上打开浏览器，阅读已登出的访客的公开帖子，这些帖子会在X、LinkedIn和Hacker News上显示，并作为结构化数据反馈给你的客服。 问你的客服“@naval和@paulg这周发布了哪些入职信息？”它会打开真正的浏览器窗口，阅读这两个时间线，并回答帖子、互动数和链接。 一切都在本地发生。没有API密钥，没有账户，没有云服务，除了页面请求本身，没有数据从你的机器流出。 会话Cookie会在每次运行前从浏览器配置文件中删除。它会读取登录访问者显示的内容——登录后没有任何信息，挑战后没有任何信息。它不解决验证码，不掩饰它是自动化的，也不会重试拒绝。 这个约束正是关键，它决定了工具能做什么、不能做什么。下面的章节对这两方面都坦诚。
概述（AI 翻译）
阅读人们实际发布的内容——在一个从未登录的浏览器中。
一个在你自己电脑上打开浏览器的MCP服务器，读取公共帖子 签出的访客会在X、LinkedIn和Hacker News上显示，并交还给他们 作为你的客服可以推理的结构化数据。
安装 ·工具 ·登出权限能为你带来什么 ·阻挡 ·安全性
这是什么
问问你的客服*“@naval和@paulg发布了哪些关于入职的帖子 一周？“* 它打开了一个真正的浏览器窗口，读取那两条时间线，然后 回答内容包括帖子、互动数和链接。
一切都在本地发生。没有API密钥，没有账户，没有云服务， 除了页面请求本身，机器上没有数据流出。
**它从未登录。**会话Cookie会从浏览器配置文件中删除 每次运行前。它读取的正是登出访客所看到的内容——什么都没有 登录后面，挑战后面什么都没有。它不能解决验证码，不能 伪装成它是自动化的，且不会重试拒绝。
这个约束就是关键，它决定了工具能做什么、不能做什么。那 下面的章节对这两者都坦诚。
快速开始
本页由 AI 翻译，技术细节请以原文为准。
README
以下为原文 README 节选，完整内容请查看原文。
What this is
Ask your agent “what have @naval and @paulg been posting about onboarding this week?” and it opens a real browser window, reads those two timelines, and answers with the posts, their engagement counts and their links.
Everything happens locally. There is no API key, no account, no cloud service and no data leaving your machine except the page requests themselves.
It never signs in. Session cookies are deleted from its browser profile before every run. It reads exactly what a logged-out visitor is shown — nothing behind a login, nothing behind a challenge. It does not solve CAPTCHAs, does not disguise that it is automated, and does not retry a refusal.
That constraint is the point, and it shapes what the tool can and cannot do. The sections below are honest about both.
Quick start
claude mcp add subio-scrape -- npx -y github:alijancb/subio-mcp
Then ask, in plain language:
собери, что @naval и @paulg писали за неделю про онбординг
Using a different agent — Codex, Cursor, Windsurf, Cline, Zed, VS Code? See INSTALL.md, which has a one-page guide per client that your agent can follow for you.
For Claude Code there is also a plugin, which adds a /scrape command and a
workflow skill on top of the tools:
git clone https://github.com/alijancb/subio-mcp.git ~/subio-mcp && cd ~/subio-mcp && npm install
claude plugin marketplace add ~/subio-mcp && claude plugin install subio-scrape@subio
Requirements
- Node.js 20 or newer.
- A browser. It uses Chrome or Edge if you have one. If not, run
npx playwright install chromiumonce. - A desktop session. The window has to be visible — see below for why. This does not run on a headless server.
Tools
| Tool | What it reads |
|---|---|
x_account_posts |
Posts from named X accounts, with exact view / like / repost / reply counts. match filters them by keyword or hashtag |
x_account_profile |
Bio, followers, post count for one account |
linkedin_page |
A public company or person page, from its JSON-LD |
linkedin_company_posts |
A company’s posts, when LinkedIn serves them signed out |
hackernews_search |
Hacker News stories and comments on a topic |
browser_status |
Which browser will be driven, from where, how fast, and what happens on a block |
browser_close |
Close the window early |
安装
npx playwright install chromium` once.