Turning PDFs into Markdown is the most common document task of the LLM era — RAG knowledge bases, fine-tuning corpora, and migrations into Notion or Obsidian all start with clean structured text. This guide covers three things: which tool to use, what quality actually means, and where the traps are — plus the fastest way to get started.

TL;DR

Scenario	Recommended approach	Why
One-off files	Online demo	Runs in your browser, nothing uploaded
Batch / scripted jobs	pdf2md CLI or Node.js / Python bindings	One command or a few lines, pipeline-ready
Highest conversion quality	pdf-inspector	#1 overall score 0.875, table TEDS 0.814

Why pdf-inspector? It's Firecrawl's open-source Rust parser. On the official opendataloader-bench (200 real PDFs) it scores 0.875 overall — first place — while processing the whole corpus in 0.470 seconds, over 30× faster than markitdown. Details on the benchmark page.

Option 1: Online conversion (zero install)

Open the home-page demo and drop in a PDF:

• Fully local: parsing runs inside your browser via WebAssembly; the file never touches a server;
• Free and unlimited: no sign-up, no page-count caps;
• Classify before converting: it tells you whether the PDF is text-based or scanned up front, so you're never confused by garbage output;
• Take results anywhere: copy the Markdown or download it as a .md file.

Note

The browser demo does not include OCR. If your file is scanned, the demo says so explicitly — that's when you need the native packages with OCR (see "What about scanned PDFs?" below).

Option 2: the pdf2md CLI (best for batches)

After installing, convert from the terminal:

cargo install pdf-inspector   # ships two CLIs: pdf2md and detect-pdf

pdf2md report.pdf -o report.md        # PDF → Markdown
detect-pdf report.pdf                 # classification only, no body output

Combine with shell loops for bulk conversions, cron jobs, and ETL pipelines. Full flag reference in the CLI guide.

Option 3: code integration (Node.js / Python / Rust)

Node.js

const { processPdf } = require('@firecrawl/pdf-inspector')

const result = await processPdf('report.pdf')
console.log(result.markdown)   // clean GitHub-Flavored Markdown

Python

from pdf_inspector import process_pdf

result = process_pdf("report.pdf")
print(result.markdown)         # Markdown string, ready to use

More APIs (region extraction, selective OCR, confidence routing) live in the quick-start guide and the API quick reference.

What "conversion quality" actually means

Plenty of tools produce Markdown-looking output; few preserve structure. Four things separate good from bad:

1. Heading levels

Many tools flatten headings into plain paragraphs, which breaks chapter-aware chunking in RAG. pdf-inspector maps font sizes to H1–H4; on the official benchmark markitdown scores 0.000 on headings while pdf-inspector scores 0.788.

2. Table recovery

Financial tables get flattened into scrambled text by weak converters. pdf-inspector combines drawn-rule detection with alignment heuristics to rebuild cell structure as standard Markdown tables, stitching continuations across pages — table TEDS 0.814, best in class.

3. Reading order

Reading a two-column paper top-to-bottom interleaves the columns. Layout analysis reconstructs true reading order (0.915 on the benchmark); RTL text is supported too.

4. Garbled glyphs and exotic fonts

Design-exported PDFs often embed CID/Type0 fonts that weaker tools render as boxes or hex soup. pdf-inspector decodes them through ToUnicode CMaps (CJK included); pages it genuinely can't decode are flagged instead of silently returning wrong text.

What about scanned PDFs?

About 54% of PDFs are native text and can be extracted fast and free; the rest are scans or image-based and need OCR. The right architecture is classify first, then route:

1. Check the type with detect-pdf or the online demo (~10–50ms);
2. TextBased → convert locally at zero OCR cost;
3. Scanned / Mixed → send only the necessary pages to OCR (the native packages ship selective OCR, tagging every page native / ocr / fused).

See Is my PDF text-based or scanned? for the details.

FAQ

Can I feed the output straight into RAG?

Yes. Output is standard GitHub-Flavored Markdown with heading levels, lists, tables, and page markers — token-efficient, and the page markers give you citation-level provenance back to specific pages.

Is there a free PDF-to-Markdown converter?

Yes — the online parser on this site is free with no sign-up and no page limits, and your file never leaves the device. The library itself is MIT-licensed and free for commercial use.

How does it compare to markitdown?

Same benchmark: overall 0.875 vs 0.589, speed 0.470s vs 16.165s, headings 0.788 vs 0.000. Full breakdown in the markitdown vs pdf-inspector comparison.

Does it handle encrypted PDFs?

Password-protected PDFs cannot be parsed directly — remove protection first. Error codes and recovery advice for damaged files are in the troubleshooting guide.

Next steps

• Try it online: drop in a PDF and see the verdict plus Markdown instantly;
• Install: npm / PyPI / crates.io — five ways in;
• Getting started: your first conversion in ten minutes.