paper2md

Precision PDF-to-Markdown converter for research papers.

Features

• Title, author, and abstract extraction from diverse paper formats
• Heading hierarchy detection via font size, weight, and allcaps analysis
• Math rendering with CM font-to-LaTeX mapping (~120 symbols)
• Tables detected via line-based layout, output in pipe format
• Figures from raster (xref), vector (drawings), and clustered composites
• References with bracket and alphanumeric key parsing
• OCR fallback for scanned PDFs (PyMuPDF OCR or pytesseract)
• Multi-column support via 1D clustering with adaptive thresholds
• MCP server with tools for PDF conversion, structured extraction, and metadata

Installation

pip install paper2md

With MCP server support:

pip install paper2md[mcp]

With OCR support for scanned PDFs:

pip install paper2md[ocr]

Usage

CLI

paper2md paper.pdf -d output/

This writes the Markdown file and all extracted figure images to the output directory.

Python API

from paper2md import convert

result = convert("paper.pdf")
print(result.markdown)

MCP Server

paper2md exposes three MCP tools: convert_pdf, convert_pdf_structured, and extract_metadata. Configure your MCP client to launch paper2md.mcp_server.

Tested Formats

paper2md is tested against papers from the following venues and publishers:

arXiv, NeurIPS, CVPR, ICLR, IEEE, ACM, NAACL, Meta AI, DeepMind, JMLR, Nature, Springer

Requirements

• Python 3.10+
• PyMuPDF 1.24+

License

MIT