research-mcp
Search and read arXiv papers directly from Claude. Supports keyword, author, category, and date filtering plus full PDF text extraction so Claude can read, summarise, and reason over entire papers, not just abstracts.
README
MCP server for searching and retrieving papers from arXiv, including full-text PDF extraction.
Features
- search_arxiv: Search papers by keyword, author, category, and/or date range
- search_papers: Quick search by query text with optional category and date filters
- get_paper: Fetch full metadata (title, authors, abstract, PDF link) for a specific arXiv ID
- get_recent: Get the most recent papers in a given category
- fetch_pdf: Download and extract full text from a paper's PDF by arXiv ID
Available Tools
| Tool | Inputs | Returns |
|---|---|---|
search_arxiv |
keyword, author, category, date_from, date_to, max_results, start |
Numbered list of papers with title, ID, authors, date, PDF link |
search_papers |
query, category, max_results, date_from |
Numbered list of matching papers with title, ID, authors, date, PDF link |
get_paper |
arxiv_id |
Title, ID, authors, published date, abstract, PDF URL |
get_recent |
category, max_results |
Numbered list of recent papers with title, ID, authors, date, PDF link |
fetch_pdf |
arxiv_id |
Full extracted text content of the paper PDF |
Usage
Clone and run
git clone https://github.com/younesbensafia/arxiv-mcp-server
cd arxiv-mcp-server
uv run arxiv-mcp-server
Test with MCP Inspector
- Run the server:
uv run arxiv-mcp-server - Open MCP Inspector or run
npx @modelcontextprotocol/inspector - Set transport to STDIO
- Command:
uv - Args:
run arxiv-mcp-server - Working directory: path to this repo
- Click Connect -- all tools appear on the left
Connect to Claude.ai
Add to your MCP settings:
{
"mcpServers": {
"arxiv": {
"command": "uv",
"args": ["run", "--directory", "/path/to/arxiv-mcp-server", "arxiv-mcp-server"]
}
}
}
arXiv Category Codes
| Code | Area |
|---|---|
cs.AI |
Artificial Intelligence |
cs.CL |
Computation and Language |
cs.CV |
Computer Vision and Pattern Recognition |
cs.IR |
Information Retrieval |
cs.LG |
Machine Learning |
cs.NE |
Neural and Evolutionary Computing |
cs.SE |
Software Engineering |
eess.AS |
Audio and Speech Processing |
math.ST |
Statistics Theory |
physics.chem-ph |
Chemical Physics |
physics.optics |
Optics |
q-bio.BM |
Biomolecules |
q-fin.ST |
Statistical Finance |
quant-ph |
Quantum Physics |
stat.ML |
Machine Learning (Statistics) |
stat.TH |
Statistics Theory |
Full list at arxiv.org/category_taxonomy.
Examples
Ask Claude (or any MCP-compatible assistant):
- "Find the latest papers on RAG from the last 3 months"
- "Summarise arxiv paper 2301.07041"
- "What are the most recent cs.LG papers today?"
- "Find all papers by Andrej Karpathy"
- "Read the full text of 1706.03762 and explain the methodology"
Don't have an MCP-compatible client? Try Claude Code or the MCP Inspector with
uv run arxiv-mcp-server.
Demo
search_papers
<p float="left"> <img src="assets/demo/search_papers/query.png" width="48%" /> <img src="assets/demo/search_papers/result.png" width="48%" /> </p>
get_paper
<p float="left"> <img src="assets/demo/get_paper/arxiv_id.png" width="48%" /> <img src="assets/demo/get_paper/result.png" width="48%" /> </p>
fetch_pdf
<p float="left"> <img src="assets/demo/fetch_pdf/arxiv_id.png" width="32%" /> <img src="assets/demo/fetch_pdf/result_1.png" width="32%" /> <img src="assets/demo/fetch_pdf/result_2.png" width="32%" /> </p>
Contributing
See CONTRIBUTING.md.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。