arXiv MCP Server
israelwcg011/arxiv-mcp-server
Acesso integrado ao arXiv para busca e gestão de papers acadêmicos
Ver no GitHubServidor MCP que expõe a API do arXiv como ferramentas e recursos para LLMs, permitindo buscar artigos, recuperar metadados, obter links de PDFs e gerar citações BibTeX diretamente em conversas com Claude. Construído com FastMCP e httpx, oferece suporte para transporte SSE e stdio.
Instalação
npx convez-ai israelwcg011/arxiv-mcp-server1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Pesquisadores buscando papers relacionados a tópicos específicos durante análise
- Acadêmicos gerando bibliografias formatadas em BibTeX automaticamente
- Cientistas recuperando metadados completos e URLs de PDFs sem sair da conversa
Veredito

Ferramenta bem estruturada e prática para integrar busca acadêmica ao workflow com Claude. Implementação limpa com FastMCP, oferece recursos documentados e suporta múltiplos transportes, atendendo bem pesquisadores e desenvolvedores que precisam automatizar tarefas bibliográficas.
README
A tradução para o português deste README ainda está na fila.
arXiv MCP Server
A Model Context Protocol (MCP) server that exposes the arXiv API as a set of tools and resources for LLMs. It allows an AI assistant (e.g. Claude Desktop) to search arXiv, retrieve paper metadata, fetch PDF links, and generate BibTeX citations directly inside a conversation.
Features
Tools
| Tool | Description |
|---|---|
search_papers | Search arXiv using the full query syntax (field prefixes, boolean operators, date filters, sorting). |
get_paper | Retrieve full metadata for a specific paper by arXiv ID. |
get_paper_pdf_url | Return the direct PDF URL for a given arXiv ID. |
create_bibtex_list_from_arxiv_ids | Generate BibTeX entries for a list of arXiv IDs. |
Resources
| Resource | Description |
|---|---|
docs://arxiv/api | Full arXiv API documentation (query syntax, field prefixes, operators, subject categories, response format) — exposed so the LLM can construct valid queries on its own. |
Running Locally
1. Clone the repo
git clone https://github.com/israelwcg011/arxiv-mcp-server.git
cd arxiv-mcp-server
2. Set up a virtual environment
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
3. Run the server
By default, server.py runs with the SSE transport on port 8080 (or the PORT environment variable, if set):
python server.py
You should see:
Starting MCP server 'arxiv' with transport 'sse' on http://0.0.0.0:8080/sse
If you prefer stdio transport (used by Claude Desktop's local-process config or MCP Inspector's default mode), edit the bottom of server.py:
if __name__ == "__main__":
# cloud / remote run
# mcp.run(transport="sse", host="0.0.0.0", port=port)
# local run
mcp.run(transport="stdio")
4. Test locally with MCP Inspector
For SSE:
npx @modelcontextprotocol/inspector --transport sse --server-url http://localhost:8080/sse
For stdio:
npx @modelcontextprotocol/inspector python server.py
5. Connect it to Claude Desktop (local)
Add to your claude_desktop_config.json:
{
"mcpServers": {
"arxiv": {
"command": "/absolute/path/to/venv/bin/python",
"args": ["/absolute/path/to/server.py"]
}
}
}
Restart Claude Desktop. You should see search_papers, get_paper, get_paper_pdf_url, and create_bibtex_list_from_arxiv_ids available as tools.
Example Queries
Search for recent papers on a topic:
search_papers(query="all:\"large language models\"", sort_by="submittedDate", sort_order="descending", max_results=10)
Field-specific search with boolean operators:
search_papers(query="au:Hinton AND cat:cs.LG")
Get metadata for a specific paper:
get_paper(arxiv_id="1706.03762")
Generate BibTeX for multiple papers:
create_bibtex_list_from_arxiv_ids(arxiv_ids=["1706.03762", "1502.03167"])
The docs://arxiv/api resource contains the full query syntax reference (field prefixes, boolean operators, date filters, subject category codes) so an LLM can construct advanced queries without external documentation.
Deployment
This server was containerized with Docker and deployed on Fly.io using:
fly launch
fly deploy
server.py reads the port from the PORT environment variable (Fly.io sets this dynamically) and supports transport="sse", host="0.0.0.0" for remote/cloud deployment. The Dockerfile and fly.toml are included in the repo as a reference for anyone wanting to deploy their own instance.
Notes
- The arXiv API returns Atom/XML; this server parses it into JSON before returning results to the client.
max_resultsis capped at 100 per the arXiv API's recommended limits.- This is a read-only integration — no data is written back to arXiv or any external service.