arxiv-mcp-server: an MCP server for searching, downloading, and analyzing scientific papers from arXiv

What it solves

Researchers and AI users often struggle to bridge the gap between LLM interfaces and the vast repository of scientific literature on arXiv. This project provides a standardized way for AI agents and clients to search, retrieve, and deeply analyze academic papers without manual downloading or copy-pasting.

How it works

This project implements the Model Context Protocol (MCP), allowing AI clients like Claude Desktop or OpenAI Codex to interact with arXiv via a set of specialized tools. It can search the arXiv API, download papers (converting them to Markdown or using PDF fallbacks), and retrieve original LaTeX source code by section. It also integrates with the Semantic Scholar API to follow citation graphs and includes an optional local semantic search feature for analyzing downloaded papers.

Who it’s for

• AI Researchers looking to automate literature reviews or research discovery.
• Developers building AI agents that require grounded access to scientific data.
• Students and Academics using LLM-powered IDEs or desktop apps to synthesize complex topics.

Highlights

• Multi-modal retrieval: Supports searching via query, downloading full text as Markdown, and retrieving specific LaTeX sections.
• Citation Intelligence: Fetches references and citing papers via Semantic Scholar.
• Advanced Workflows: Includes built-in prompt workflows for deep paper analysis, literature synthesis, and research question formulation.
• Local-First Storage: Papers and semantic indexes are stored locally on the user's machine for privacy and speed.