arXiv MCP Server

Content

arXiv MCP Server connects MCP clients to arXiv research workflows. It can search papers, download paper content, read locally downloaded papers, list local papers, run semantic search over the local collection, fetch citation graph data, save research alerts, and provide research-oriented prompts.

The upstream README documents a portable uvx arxiv-mcp-server MCP configuration. The PyPI package exposes the arxiv-mcp-server command for MCP clients.

Source Review

• https://github.com/blazickjp/arxiv-mcp-server
• https://github.com/blazickjp/arxiv-mcp-server/blob/main/README.md
• https://github.com/blazickjp/arxiv-mcp-server/blob/main/pyproject.toml
• https://pypi.org/pypi/arxiv-mcp-server/json

These sources were reviewed on 2026-06-05. Prefer the live repository and PyPI metadata for current package version, command name, Python requirement, optional extras, tool behavior, security guidance, and setup guidance.

Features

• Search arXiv papers with optional category, date, and boolean filters.
• Download papers and store them locally for later reading.
• Read downloaded paper text in bounded chunks for large papers.
• List papers already downloaded into the local collection.
• Run semantic search over locally downloaded papers when pro dependencies are installed.
• Fetch references and citing papers for arXiv IDs.
• Save topic watches and poll for newly published papers.
• Use research prompts for summaries, comparisons, literature review, and paper analysis.

Installation

For MCP clients that launch stdio servers:

{
  "mcpServers": {
    "arxiv": {
      "command": "uvx",
      "args": ["arxiv-mcp-server"]
    }
  }
}

Restart the MCP client after adding the server.

Use Cases

• Ask Claude to search arXiv for recent papers on a topic.
• Download a paper, read it in chunks, and summarize the core argument.
• Compare several papers already saved in the local collection.
• Search locally downloaded papers for related methods or terminology.
• Explore references and citing papers around an arXiv ID.
• Set watches for research topics and review newly published papers.

Safety and Privacy

Paper content is untrusted external input. A malicious or adversarial paper can contain prompt-injection text that tries to override instructions or trigger unintended tool use. Treat paper text and model summaries as data, require human review before acting on paper-derived instructions, and be especially cautious when this server is used alongside tools that can write files, run commands, send messages, browse sites, or access private systems.

Research workflows can reveal sensitive direction. Queries, downloaded papers, paper IDs, citation graph requests, semantic search terms, local storage choices, and generated summaries may expose private research, product, academic, legal, or competitive-analysis context. Review local paper storage and generated notes before sharing or syncing them.

Duplicate Check

No blazickjp/arxiv-mcp-server entry, arxiv-mcp-server package entry, or matching source URL was found in content/mcp.