Convert PDF to Markdown for RAG & LLMs

Feeding raw PDF text to an LLM throws away the structure your retrieval depends on. PDFjet returns clean Markdown — headings, lists and tables preserved in reading order — so you can chunk on real document structure and get better answers out of your RAG pipeline.

1. curl

curl -X POST https://pdfjet.dev/extract/md \
  -H "Authorization: Bearer pj_live_..." \
  -F [email protected] \
  -o paper.md

2. Python (into your vector store)

import requests

md = requests.post(
    "https://pdfjet.dev/extract/md",
    headers={"Authorization": "Bearer pj_live_..."},
    files={"file": open("paper.pdf", "rb")},
).text

# chunk on headings, then embed
chunks = [c for c in md.split("\n## ") if c.strip()]

3. Straight from your AI assistant (MCP)

PDFjet ships an MCP server, so Claude, Cursor and other MCP clients can convert a PDF to Markdown with no glue code:

{
  "mcpServers": {
    "pdfjet": {
      "command": "npx",
      "args": ["-y", "pdfjet-mcp"],
      "env": { "PDFJET_API_KEY": "pj_live_..." }
    }
  }
}

Why Markdown beats a text dump

Flat pdftotext output merges columns, drops tables into noise, and loses heading boundaries — so your chunks straddle unrelated sections and retrieval degrades. Structured Markdown lets you split on #/## boundaries and keeps tables as tables, which the model can actually read.

See it on your own file first — preview the Markdown in the free tool, no signup.