modou-tools-mcp
modou-tools-mcp
给中文开发者的全球技术情报台。
一个 MCP server,装进 Agent 就能采集全球技术情报:英文侧看 GitHub 趋势、arXiv 论文、HN 热帖、Exa 语义搜索、SEC 美股财报;中文侧深入掘金、知乎、B站、A股财报。全部工具 零 API key、零费用、国内直连——这是它跟 Exa / Firecrawl / Perplexity 这类要订阅要 key 的方案最大的区别。
快速开始(云端托管版)
在支持 MCP 的客户端(Claude Code / Cherry Studio / Kimi Playground 等)中添加如下配置即可连接,无需本地安装任何依赖:
{
"mcpServers": {
"modou-tools": {
"command": "uvx",
"args": ["modou-tools-mcp@latest"]
}
}
}Related MCP server: Agent中国通
本地部署(完整功能)
云端托管版为兼容性考虑不包含本地搜索(见已知限制)。需要完整功能时,本地运行:
# 本机 Python 环境(建议 3.10+)
pip install modou-tools-mcp{
"mcpServers": {
"modou-tools": {
"command": "uvx",
"args": ["modou-tools-mcp@latest"]
}
}
}或使用仓库源码直接运行(开发模式):
git clone https://github.com/xihongshi567/modou-tools-mcp
cd modou-tools-mcp
pip install -e .
claude mcp add modou-tools -- python "D:/okok/modou-tools-mcp/src/modou_tools_mcp/server.py"工具清单
工具 | 用途 | 网络 |
| 中文网页搜索(SearXNG 聚合,仅本地部署可用) | 本地 |
| 网页正文提取,去噪输出 Markdown(trafilatura) | 直连 |
| 热榜聚合(知乎/B站/GitHub/HN) | 直连 |
| GitHub 仓库趋势/对比(60次/h,配 token 5000次/h) | 直连 |
| arXiv 论文检索(支持分类/高级语法,按相关度或最新排序) | 直连 |
| 英文语义搜索(Exa 免费端点,免 key) | 直连 |
| 掘金技术社区文章发现(推荐流/分类浏览) | 直连 |
| 美股财报(SEC EDGAR 免费) | 可用,较慢 |
| A股财报(东方财富公开接口) | 直连 |
| PDF 报告 → Markdown + 表格(PyMuPDF) | 本地 |
所有工具返回统一结构:成功 {"ok": true, ...};失败 {"ok": false, "error": "...", "detail": "..."}。
一个完整场景:调研"RAG 技术方向"
Agent 拿到一个调研任务,modou-tools 全链路支撑:
开题:
exa_search("retrieval augmented generation trends 2026")摸全球风向补学术:
arxiv_search("cat:cs.CL AND all:RAG")看最新论文看社区:
juejin_recommended/fetch_hot_topics(zhihu)看国内在聊什么抓原文:
fetch_page(url)抓关键文章正文读报告:
parse_pdf(url)解析 PDF 白皮书查数据:
github_search_repos("RAG framework")看项目热度
设计原则
纯采集:不做 LLM 加工(Agent 本身会总结提炼),server 保持轻量
零依赖:除可选 GITHUB_TOKEN 外不需要任何 key;Exa 走官方免费端点
错误不打断:单工具失败返回结构化错误,Agent 可换路重试
环境变量
变量 | 说明 | 必填 |
| GitHub API token,提升速率限制至 5000 次/h | 可选 |
开发
# 离线冒烟(结构检查,不依赖网络)
python smoke_test.py
# 在线验证全部工具
python scripts/verify_online.py已知限制
web_search依赖本地 SearXNG(docker start searxng),云端托管版该工具不可用GitHub 无 token 仅 60 次/h(配额耗尽时返回 403 提示);SEC 偶发 503 可重试;东方财富接口偶发断连(内置重试 3 次)
Exa 免费端点有配额限制,大量调用时可能受限
掘金站内搜索接口需登录态,未提供;掘金文章全文需配合
fetch_page扫描版 PDF 返回
is_scanned=true警告(无 OCR)
Available Tools
15 toolsarxiv_searchArxiv SearchB
搜索 arXiv 论文(标题/摘要/作者全字段匹配),研究者获取最新学术动态。
Inputs: query (required), sort, limit. Returns: object
cninfo_company_profileCninfo Company ProfileB
查询 A股公司基本信息(东方财富公开接口)。
Inputs: code (required), market. Returns: object
cninfo_compare_companiesCninfo Compare CompaniesA
对比多家 A股公司最新财年核心指标。
Inputs: codes (required), years. Returns: object
cninfo_financial_analysisCninfo Financial AnalysisA
查询 A股公司财务分析指标(ROE/净利率/营收增速/毛利率/研发占比)。
Inputs: code (required), years. Returns: object
exa_searchExa SearchC
Exa 英文语义搜索,适合海外技术趋势、行业报告、论文。
Inputs: query (required), num_results. Returns: object
fetch_hot_topicsFetch Hot TopicsA
获取主流平台当前热搜榜单(zhihu 知乎 / bilibili B站 / github GitHub 趋势 / hackernews HN)。
Inputs: platforms. Returns: object
fetch_pageFetch PageA
抓取网页正文,自动去广告/导航/侧边栏,输出干净 Markdown(trafilatura)。
Inputs: url (required), max_chars, include_links. Returns: object
github_compare_reposGithub Compare ReposB
对比多个 GitHub 仓库的当前快照数据。
Inputs: repos (required). Returns: object
github_search_reposGithub Search ReposA
搜索 GitHub 仓库趋势,把技术趋势翻译成数字(stars/forks/活跃度)。
Inputs: query (required), sort, limit. Returns: object
juejin_by_categoryJuejin By CategoryB
按分类浏览掘金文章(ai / frontend / backend / android / ios / freebie / article)。
Inputs: category (required), limit. Returns: object
juejin_recommendedJuejin RecommendedC
掘金技术社区推荐流文章发现。
Inputs: sort, limit, cursor. Returns: object
parse_pdfParse PdfA
解析 PDF 报告为 Markdown(保留表格),本地文件或在线 URL 均可。
Inputs: url, file_path, max_pages, extract_tables. Returns: object
sec_company_factsSec Company FactsB
查询美股公司财报(SEC EDGAR),拿营收/净利/研发支出等结构化数据。
Inputs: cik, years, ticker, metrics. Returns: object
sec_compare_companiesSec Compare CompaniesA
对比多家美股公司同一财年的核心指标。
Inputs: tickers (required), years, metrics. Returns: object
web_searchWeb SearchB
中文网页搜索,聚合百度/搜狗/Bing CN 等引擎(依赖本地 SearXNG localhost:8080)。
Inputs: query (required), language, num_results. Returns: object
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 15 tools
Tools are mostly distinct by source and purpose (search vs. compare vs. fetch vs. parse), but a few could be confused: web_search vs exa_search both handle search, and juejin_recommended vs juejin_by_category both return Juejin articles.
All names use snake_casecars but follow inconsistent patterns: some are source_verb_object (github_search_repos), some are verb_object (fetch_page), and some are source_attribute (juejin_recommended, cninfo_company_profile). The naming is readable but not fully predictable.
15 tools is at the upper boundary of a reasonable scope for a research/investment aggregation servercars. Each tool maps to a distinct data source or operation, though the count is getting close to heavy.
Covers searching, fetching content, financial data, and comparison workflows across multiple sources. Minor gaps exist (e.g., no explicit historical data beyond current snapshots, no saved/watchlist functionality), but core research workflows are complete.
Maintenance
Related MCP Connectors
One MCP server for all web data. Scrape, search, crawl, and AI research in one install.
Web search API and MCP server for developers building AI agents.
Driflyte MCP server which lets AI assistants query topic-specific knowledge from web and GitHub.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceAn MCP server optimized for Chinese users that enables AI models to perform Bing searches and fetch webpage content without requiring an API key. It provides specialized tools for searching and scraping web data directly through supported AI environments like Claude or Cursor.22 npm7MIT
- AlicenseAqualityDmaintenance这是一个面向中文圈的MCP服务器,将中国互联网常用能力(如地图、快递、RSS、B站等)封装为标准MCP工具,方便AI Agent安全调用。132MIT
- AlicenseNot gradedqualityDmaintenanceMCP server that aggregates AI news, academic papers from ArXiv, and trending GitHub repositories from multiple sources, enabling users to fetch, search, and filter recent AI content via natural language.1MIT
- FlicenseNot gradedqualityDmaintenanceThis MCP server enables AI agents to read and search the internet without paid API keys, supporting multiple channels like web, GitHub, RSS, YouTube, and more, with automatic browser escalation and caching.-