mcp-deep-research

🔍 McpDeepResearch

An MCP (Model-Context-Protocol) Server for Deep Academic Research

一个用于深度学术研究的 MCP 服务器

---

English | 简体中文

---

English

McpDeepResearch is a lightweight but powerful MCP (Model-Context-Protocol) server that helps you quickly discover, retrieve, and read academic papers from the web using the familiar Google Scholar interface.

💡 Reuses your local browser session

The server never fetches pages directly: it drives the locally installed Chrome over the DevTools Protocol. Every search and every fetch therefore runs with the session already present in that browser, so your logins are honoured and the access rights of your institution's literature databases (subscriptions, campus-network licences) are reused as they are — no extra credentials or proxy setup required.

✨ Features

• search_scholar_papers – Google Scholar search with optional year-filter & date-sort
• fetch_md – Convert any public web page to clean Markdown
• fetch_paper – Auto-detect the paper content (title, abstract, body, references) and strip the rest

🛠️ Prerequisites

• Python ≥ 3.10
• Google Chrome/Chromium (for headless fetching via Chrome DevTools Protocol)
• Environment variables

  export CDP_ENDPOINT="http://localhost:9222"   # Chrome debugging port
  export GOOGLE_PROXY="http://proxy:port"        # (optional) HTTP(S) proxy
  

⚙️ Quick Start

1. Install via pip

   pip install mcp-deep-research
   

2. (Optional) Manually launch Chrome if it isn’t already running
   The server will automatically launch a Chrome instance; if not successfully launched, you can launch it manually with:

   google-chrome --remote-debugging-port=9222 --user-data-dir=/tmp/chrome-profile &
   

3. Run the MCP server

   # Launch with SSE
   mcp-deep-research --transport sse --port 8000
   # Launch with STDIO
   mcp-deep-research --transport stdio
   

   The server exposes 3 read-only tools to any MCP-capable client.

🧑‍🎓 Example Workflow in a Chat-UI

1. Search
   “Find recent papers on diffusion models after 2022.”
   → search_scholar_papers("diffusion models", year=2022, sort_bd=True)

2. Fetch
   Pick an interesting PDF link from the results and call
   fetch_paper("https://arxiv.org/abs/2304.12345")

3. Read
   The cleaned Markdown (title, abstract, full text) appears directly in the chat.

🔒 Security

• 100 % read-only; no writes, no uploads, no local file access.
• All traffic respects the original site’s robots.txt.
• Proxies can be configured to stay within institutional or regional firewalls.

🤝 Contributing

PRs are welcome!

---

简体中文

McpDeepResearch 是一个轻量级、但功能完备的 MCP（Model-Context-Protocol）服务器，帮助你在 Google Scholar 上快速发现、抓取并阅读学术文献。

💡 复用本地浏览器会话

服务器不直接请求网页，而是通过 DevTools Protocol 驱动本机已安装的 Chrome。因此每一次检索与抓取都沿用该浏览器中已有的会话：登录状态被正常识别，所属机构文献数据库的访问权限（订阅、校园网许可等）可直接复用，无需额外配置凭据或代理。

✨ 功能一览

• search_scholar_papers – 使用关键词在 Google Scholar 中搜索，可过滤年份 / 按日期排序
• fetch_md – 将任意公开网页渲染为整洁的 Markdown
• fetch_paper – 智能提取网页中的论文主体，去除广告、导航条等噪声

🛠️ 前置条件

• Python ≥ 3.10
• Google Chrome / Chromium（通过 CDP 进行无头抓取）
• 环境变量

  export CDP_ENDPOINT="http://localhost:9222"   # Chrome 调试端口
  export GOOGLE_PROXY="http://proxy:port"        # 可选：HTTP(S) 代理
  

⚙️ 快速开始

1. 通过 pip 安装

   pip install mcp-deep-research
   

2. （可选）如果 Chrome 尚未启动可手动启动
   服务器启动时会自动运行 Chrome 实例。若未成功启动，可手动启动：

   google-chrome --remote-debugging-port=9222 --user-data-dir=/tmp/chrome-profile &
   

3. 启动 MCP 服务器

   # 以 SSE 启动
   mcp-deep-research --transport sse --port 8000
   # 以 STDIO 启动
   mcp-deep-research --transport stdio
   

   服务器会对外暴露 3 个只读工具。

🧑‍🎓 对话界面中的典型工作流

1. 搜索
   “找 2022 年之后关于扩散模型的论文。”
   → search_scholar_papers("diffusion models", year=2022, sort_bd=True)

2. 抓取
   从结果中挑选一篇 PDF 链接，调用
   fetch_paper("https://arxiv.org/abs/2304.12345")

3. 阅读
   清洗后的 Markdown（含标题、摘要、全文）直接展示在聊天窗口。

🔒 安全性

• 完全只读，不修改、不上传、不写入本地文件。
• 所有请求均尊重目标站点的 robots.txt。
• 可配置代理以符合校园网或公司网络的安全策略。

🤝 如何贡献

欢迎提 PR！

---