Open Source · Model Context Protocol

Your AI Research
Assistant for ArXiv

Search, read, and reason over scientific papers. Powered by MCP, ChromaDB, and local embeddings — no API keys required.

$ git clone https://github.com/rishimule/arxiv-intelligence-mcp.git && cd arxiv-intelligence-mcp && uv sync

Everything you need for
AI-powered research

Connect to Claude Desktop or Cursor and start exploring the scientific literature with natural language.

Smart Search

Query ArXiv's entire database with natural language. Results are ranked by relevance with full metadata.

Deep Ingestion

Download PDFs and convert them to clean Markdown using IBM DocLing. Every figure and table preserved.

Semantic Memory

Papers are chunked and embedded locally with sentence-transformers. No API keys, no data leaves your machine.

RAG Q&A

Ask questions about your ingested library. Answers are grounded in actual paper content with cited sources.

How it works

A clean pipeline from your AI assistant to the scientific literature and back.

Claude Desktop / Cursor
MCP Protocol
ArXiv Intelligence Server
ArXiv API — search
DocLing — PDF to Markdown
ChromaDB — vectors

Tools & Resources

Four endpoints that give your AI assistant full research capabilities.

Type Name Description
Tool search_arxiv Search ArXiv papers by query, sorted by relevance
Tool ingest_paper Download, process, and embed a paper into local storage
Tool ask_paper RAG query across your ingested paper library
Resource research://library List all papers currently in your library

From question to answer

A natural four-step research workflow, driven entirely by conversation.

1

Discover

"Find the top 3 papers on vision transformers"

2

Ingest

"Read the second paper and add it to my library"

3

Query

"Explain the architecture described in that paper"

4

Synthesize

"Compare this approach to standard transformers"

Up and running in minutes

Three commands. No API keys. Fully local embeddings.

bash
# Clone and install
$ git clone https://github.com/rishimule/arxiv-intelligence-mcp.git
$ cd arxiv-intelligence-mcp
$ uv sync

# (Optional) Copy environment config
$ cp .env.example .env

# Start the MCP server
$ uv run arxiv-intelligence-mcp
claude_desktop_config.json
{
  "mcpServers": {
    "arxiv-intelligence": {
      "command": "/path/to/.venv/bin/python",
      "args": ["-m", "arxiv_mcp.main"],
      "cwd": "/path/to/arxiv-intelligence-mcp",
      "env": { "PYTHONPATH": "src" }
    }
  }
}

Tech Stack

Modern Python tooling with local-first architecture.

FastMCP
MCP Framework
DocLing
PDF Conversion (IBM)
ChromaDB
Vector Database
sentence-transformers
Local Embeddings
arxiv.py
ArXiv API Client
uv
Package Manager