Choosing a PDF to markdown converter matters more than most people realize. A PDF stores positioned text, not a document outline, so every tool has to decide how hard it tries to reconstruct headings, lists and tables. Some only pull out the text. Some rebuild the structure with layout analysis or machine-learning models. Some run OCR on scans. The right one depends on what your PDFs look like and where the output is going.
We rechecked every tool below in October 2026 against its own website, README or pricing page, and against what Craft Markdown's converter code does. Here is what we found.
Quick Comparison Table
| Tool | Where it runs | Free | Structure from PDFs | OCR | Other formats | Best For |
|---|---|---|---|---|---|---|
| Craft Markdown | Your browser | Unlimited | Headings from font size, lists, bold, simple pipe tables | No | 11 more (Word, Excel, HTML, CSV, JSON, XML, TXT, RTF, MD) | Fast, private conversion; multi-format |
| pdf2md.morethan.io | Your browser | Yes | Headings, lists, bold and italic from font heuristics | No | PDF only | Quick single PDFs with some structure |
| MarkItDown (Microsoft) | Python, local | Yes (MIT) | Plain text; tables via plugins | Via plugin or Azure | 15+ | Python pipelines, LLM prep |
| Docling (IBM) | Python, local | Yes (MIT) | Headings, tables, reading order via ML models | Yes, built in | 20+ | Best open-source structure recovery |
| Pandoc + pdftotext | Command line, local | Yes | Plain text (Pandoc cannot read PDF itself) | No | 40+ into Pandoc | Scripted pipelines already on Pandoc |
| LlamaParse | Cloud API | 10,000 credits/month free | Headings, tables, multi-column via LLMs | Yes | 130+ | Complex documents in production RAG |
| Adobe Acrobat | Desktop and cloud | No ($19.99/mo Pro) | Exports to Word or HTML, not Markdown | Yes, best in class | Many | OCR on scans, then convert the export |
How We Evaluated These Tools
- Privacy — Does the tool process files on your device, or upload them? For confidential documents this decides the list.
- Structure recovery — Does the Markdown get real headings, lists and pipe tables, or just the text in reading order?
- OCR — Can it read scanned PDFs, which are pictures of text?
- Ease of use — Drop a file, or install Python and read a README?
- Format support — PDF only, or the rest of your documents too?
- Cost — Free, free tier, or subscription?
1. Craft Markdown — Fastest Private Conversion
Overview
Craft Markdown is a browser-based converter. Drop a PDF and PDF.js, running in your tab, reads the text layer of every page. The converter then rejoins wrapped lines into paragraphs, turns clearly larger fonts into #, ## and ### headings, picks up bullet and numbered lists, bold runs and simple aligned tables, and drops running headers, footers and page numbers that repeat on every page. Nothing is uploaded. The same page converts Word, Excel, HTML, CSV, JSON, XML, text and RTF files.
Pros
- Nothing leaves your device: the conversion is JavaScript in your browser. There is no server to receive the file.
- No setup: no signup, no install, no Python. Drop a file, download the .md.
- Clean text: wrapped lines are rejoined, hyphenated words are mended, repeating headers and footers are dropped, and control characters and zero-width marks are stripped, so the file opens cleanly anywhere and pastes well into ChatGPT or Claude.
- Many files at once: drop a folder's worth and download them as a ZIP.
- Multi-format: one page for 12 file types.
Cons
- Structure is heuristic. Headings come from font size (three levels; a heading that is only bold becomes a bold line), and a table becomes a pipe table only when three or more consecutive lines share the same aligned columns. Tables with merged cells, wrapped cell text or multi-line headers come out as plain text in reading order. If layout recovery fails, you get the plain text of each page.
- No OCR. Scanned PDFs produce nothing; run them through an OCR tool first. Images are not extracted.
- Unusual layouts (text boxes, multi-column) can come out in the wrong order.
- No API.
Best For
- Getting the text out of a text-based PDF quickly and privately
- Feeding a PDF's text to an AI assistant or a RAG pipeline where you chunk by heading or paragraph
- People who also need Word, Excel or HTML converted on the same page
Pricing
Free, no limits, no account. Files up to 50 MB.
2. pdf2md.morethan.io — Open-Source Browser Tool With Heading Detection
Overview
pdf2md is the hosted version of jzillmann/pdf-to-markdown, an MIT-licensed JavaScript tool. Like Craft Markdown it runs PDF.js in your browser (the site's bundle ships the PDF.js worker), so your file is not uploaded. Like Craft Markdown it goes a step further than plain extraction and tries to turn larger fonts into Markdown headings, detect lists, keep bold and italic, and drop repeating headers and footers. It does not build tables.
Pros
- Runs in the browser; no upload
- Attempts headings and lists from font sizes, which plain text extractors such as pdftotext skip
- Open source, so you can read exactly what it does
- Free, no signup
Cons
- PDF only
- Heading detection is a heuristic; unusual layouts produce odd results
- The project has had little development since its 2020 to 2021 restructuring, so bugs are unlikely to be fixed
- No OCR, no tables, no batch download
Best For
- A quick single PDF where you want a first pass at headings
Pricing
Free.
3. MarkItDown (Microsoft) — Best for Python Pipelines
Overview
MarkItDown is Microsoft's MIT-licensed Python library and CLI for converting files to Markdown for LLM use. It needs Python 3.10 to 3.14. For PDFs it extracts text with pdfminer, so the default output is plain text with no headings or tables; the optional markitdown-ocr plugin and Azure Document Intelligence add OCR and layout analysis. It also ships an MCP server so AI agents can call it as a tool. There is no official hosted version; if you want the same result without Python, see our MarkItDown online page.
Pros
- Free, open source, backed by Microsoft and actively maintained
- Handles PDF, Word, PowerPoint, Excel, HTML, images, audio, ZIP, EPUB, YouTube URLs and more
- One function call in Python, or
markitdown file.pdf -o file.mdon the command line - OCR available through a plugin or Azure
- MCP server for Claude Desktop and similar clients
Cons
- Requires Python and a pip install
- Default PDF output is plain text, without the heading, list and table detection that Craft Markdown and pdf2md attempt
- Azure options send the document to the cloud and are billed
Best For
- Developers converting many files in a script or ingestion pipeline
- Agent workflows that need a convert-to-markdown tool
Pricing
Free (MIT). Azure features are paid per call.
4. Docling (IBM) — Best Open-Source Structure Recovery
Overview
Docling started at IBM Research and is now hosted by the LF AI & Data Foundation. It is an MIT-licensed Python package (pip install docling, Python 3.10 or newer) that runs machine-learning layout and table-structure models on each page, then exports Markdown, HTML or lossless JSON. Of the free tools here it does the most to rebuild headings, reading order and tables from a PDF, and it has OCR built in for scans. It integrates with LangChain, LlamaIndex, Haystack and CrewAI.
Pros
- Recovers headings, lists, tables and reading order, including multi-column pages
- OCR for scanned PDFs and images, with no cloud call
- Runs locally, which suits sensitive data and air-gapped setups
- Reads PDF, DOCX, PPTX, XLSX, HTML, images, audio and more
- Free, MIT licensed, large community (68,000+ GitHub stars at the time of checking)
Cons
- Python only; the first run downloads model weights
- Slower than text extraction, and a GPU helps on long documents
- More moving parts than a drop-a-file tool
Best For
- RAG pipelines where tables and section structure matter
- Scanned or complex PDFs when you cannot send them to a cloud service
Pricing
Free (MIT).
5. Pandoc with pdftotext — For Pipelines Already on Pandoc
Overview
Pandoc is the standard command-line document converter, but it has no PDF reader: pandoc file.pdf fails. The usual pairing is Poppler's pdftotext to get the text, then Pandoc if you need to go on to another format:
pdftotext -layout report.pdf report.txt
That gives you plain text, without the headings, lists and simple tables Craft Markdown tries to recover. Pandoc earns its place when the rest of your pipeline is Pandoc already, for example turning the text into DOCX or HTML afterwards, or when the source is DOCX, HTML, EPUB or one of its 40+ real input formats.
Pros
- Scriptable; converts thousands of files unattended
- Local processing
- Unmatched for DOCX, HTML, LaTeX and Markdown-to-anything conversion
Cons
- Does not read PDF. You need pdftotext or another extractor first.
- Command line only, with a learning curve
- The result is plain text, not structured Markdown
Best For
- Batch jobs and CI pipelines that already use Pandoc
Pricing
Free, open source (GPL).
6. LlamaParse — Best Hosted Parser for Complex Documents
Overview
LlamaParse is LlamaIndex's cloud parsing API. It uses layout models and LLMs to rebuild headings, tables and multi-column text, and it reads 130+ file types with OCR. It is the tool to reach for when Docling's output is not good enough and the documents can leave your network.
Pros
- Strongest results on complex tables, forms and multi-column layouts
- OCR included
- Native LlamaIndex integration
Cons
- Documents are uploaded to the cloud
- Usage based: credits per page, more for the agentic modes
- API integration, not a drop zone
Best For
- Production RAG over complex documents, with budget
Pricing
Free plan with 10,000 credits a month (roughly 3,300 pages at the cost-effective tier, 10,000 at the fast tier); Starter $50/month for 40,000 credits; Pro $500/month. Checked on llamaindex.ai/pricing, October 2026.
7. Adobe Acrobat — Best OCR, No Markdown Export
Overview
Acrobat Pro remains the reference for PDF editing and OCR, but it does not export Markdown. Its route to Markdown is indirect: run OCR in Acrobat, export to Word or HTML, then convert that with any tool above (the Word to Markdown converter keeps headings, lists and links from the Word export).
Pros
- Best-in-class OCR and layout recognition
- Fixes damaged or scanned PDFs before conversion
- Enterprise features and support
Cons
- Subscription: Acrobat Pro is $19.99/month, Standard $14.99/month on annual plans (adobe.com, October 2026)
- Markdown is not an export format
- Cloud features process documents on Adobe's servers
Best For
- Scanned documents that need high-quality OCR before anything else
Pricing
Acrobat Standard $14.99/month, Pro $19.99/month (annual, billed monthly).
Comparison by Use Case
Best for Privacy
- Craft Markdown and pdf2md — both run in your browser, nothing uploaded
- Docling and MarkItDown — local Python, no network calls in the default setup
- Pandoc + pdftotext — local command line
LlamaParse and Acrobat's cloud features send your document to a server.
Best for Structure (Headings and Tables)
- LlamaParse — strongest on complex layouts, cloud
- Docling — best open-source option, local
- Craft Markdown and pdf2md — heading, list and bold heuristics in the browser; Craft Markdown also builds simple tables
MarkItDown's default and pdftotext give plain text. Craft Markdown's headings and tables are heuristics that work on clean, single-column documents; for tables with merged cells or wrapped text you need Docling or LlamaParse.
Best for Scanned PDFs
- Adobe Acrobat — OCR, then export and convert
- Docling — OCR built in, local
- LlamaParse — OCR in the cloud
Best for Beginners
- Craft Markdown — drop a file
- pdf2md — drop a file
- Adobe Acrobat — familiar desktop app
Best for Developers
- Docling — structure recovery in Python
- MarkItDown — simplest Python API, MCP server
- Pandoc + pdftotext — shell pipelines
How to Choose
- Choose Craft Markdown when you want a text-based PDF converted now, privately, with headings, lists and simple tables picked up, and you may also have Word, Excel or HTML files to convert.
- Choose pdf2md for a quick single PDF when its heading guesses suit the document better.
- Choose MarkItDown when you are writing Python and want one call per file, or an MCP tool for an agent.
- Choose Docling when headings and tables matter and the files must stay local.
- Choose Pandoc + pdftotext when the rest of your pipeline is already Pandoc.
- Choose LlamaParse for complex documents in production, with budget and no restriction on uploading.
- Choose Adobe Acrobat when the PDFs are scans and you need the best OCR first.
Our Verdict
There is no single best PDF to markdown converter, because PDFs differ so much. For text-based PDFs with a straightforward layout, Craft Markdown is the fastest private route: drop the file, download the Markdown with headings, lists and simple tables, nothing uploaded. For PDFs where complex tables or multi-column layouts carry the meaning, Docling is the best free tool and LlamaParse the best hosted one. For scans, OCR first, with Acrobat or Docling.
Convert Your PDFs Now — Free & Private →
Frequently Asked Questions
Which PDF to markdown converter is most accurate?
For recovering structure, LlamaParse and Docling lead, because they run layout models rather than reading the text layer in order. For plain text from a clean, single-column PDF, every tool here gives similar results. No tool handles scanned PDFs without OCR.
Are these tools really free?
Craft Markdown, pdf2md, MarkItDown, Docling and Pandoc with pdftotext are free with no usage limits. LlamaParse has a free plan of 10,000 credits a month, then paid plans from $50/month. Adobe Acrobat is a subscription from $14.99/month.
Which tool is best for AI and RAG?
If the PDFs are simple, Craft Markdown's output (headings from font size, lists, simple tables, repeating headers and footers dropped) is enough and keeps the documents private. If you need complex tables, multi-column layouts or OCR, use Docling locally or LlamaParse in the cloud. MarkItDown fits Python pipelines and agent tools. See our RAG guide.
Can I convert scanned PDFs to markdown?
Only with OCR. Docling and LlamaParse include it, MarkItDown has an OCR plugin, and Adobe Acrobat has the best OCR of all. Craft Markdown, pdf2md and pdftotext read the text layer only, so a scan comes out empty. You can also OCR with Tesseract first and then convert the result with any tool.
Do any of these tools work offline?
Docling, MarkItDown and Pandoc with pdftotext run fully offline once installed. Craft Markdown and pdf2md need the page to load but then do not upload your file. LlamaParse and Acrobat's cloud features need an internet connection.
Does Craft Markdown keep tables from PDFs?
Simple ones. When three or more consecutive lines share the same aligned columns, they become a pipe table with the first row as the header. Tables with merged cells, wrapped cell text or multi-line headers come out as text in reading order. For those, convert the original DOCX, HTML or a CSV export instead, or use Docling or LlamaParse on the PDF.