Document to Markdown – PDF, Word, PowerPoint & Excel to Markdown (with OCR)

Convert PDF to Markdown, Word (DOCX) to Markdown, PowerPoint to text, Excel to Markdown, plus EPUB, HTML, CSV and images, from any public URL. You get clean, LLM-ready Markdown with headings, lists and tables, page numbers, optional RAG chunks, and built-in PDF OCR for scanned documents. It runs on Microsoft's open-source MarkItDown library, so you can use it as a hosted MarkItDown API without running any servers.

$1.50 per 1,000 pages (+ $5 per 1,000 pages that need OCR). You pay only for results: files that fail to download or contain no text are free.

Why this document parser

• ✅ One Actor for all office formats: PDF, DOCX, PPTX, XLSX/XLS, EPUB, HTML, CSV, JSON, TXT, PNG/JPG/TIFF/WebP
• ✅ Tables stay tables: PDF, Word, PowerPoint and Excel tables come out as Markdown tables
• ✅ OCR only where needed: scanned pages are detected automatically and OCRed; normal pages are not slowed down or billed for OCR
• ✅ Page control: convert only 1-5 or 10-, and pay only for those pages
• ✅ Built for RAG and LLMs: page markers like <!-- Page 3 --> for citations, optional paragraph-aware chunks with page ranges
• ✅ Per-page pricing: a 2-page invoice costs less than a 300-page report
• ✅ Share links work: Google Drive, Google Docs/Sheets/Slides, Dropbox, GitHub and arXiv links are turned into direct downloads

What you get

One item per document:

Field	Description
url	The document URL you sent
status	success, no_text (nothing readable found, not charged) or failed
fileType	pdf, docx, pptx, xlsx, xls, epub, html, csv, png, jpg, …
title	Document title from the file's metadata or first page, when available
pageCount	Pages in the document (slides for PPTX, sheets for Excel)
pagesProcessed	Pages converted in this run (after your pages selection)
ocrPages	Pages whose text came from OCR
billedPages	Pages charged (see pricing below)
wordCount	Number of words in the Markdown
markdown	The full document as Markdown
chunks	Optional [{index, text, pageStart, pageEnd}] chunks for embeddings
error	Why a document failed, in plain words

Inputs

• Document URLs (required): public links to the files. Max 50 MB per file.
• Pages: e.g. 1-5, 1,3,7-9 or 10-. Applies to PDF pages, PowerPoint slides, Excel sheets and multi-page TIFFs. Empty = everything.
• OCR: auto (default, OCR only for scanned pages and images), always (for PDFs with a broken text layer) or never.
• RAG chunk size: split the Markdown into chunks of up to N characters (0 = off).
• Mark page breaks: adds <!-- Page N --> comments (on by default).

{
  "documentUrls": ["https://arxiv.org/pdf/1706.03762"],
  "pages": "1-3",
  "ocr": "auto",
  "outputChunks": 1500
}

Example output

{
  "url": "https://arxiv.org/pdf/1706.03762",
  "status": "success",
  "fileType": "pdf",
  "title": "Attention Is All You Need",
  "pageCount": 15,
  "pagesProcessed": 3,
  "ocrPages": 0,
  "billedPages": 3,
  "wordCount": 1393,
  "markdown": "<!-- Page 1 -->\n\nAttention Is All You Need\n\nAshish Vaswani ...\n\nAbstract\n\nThe dominant sequence transduction models ...",
  "chunks": [{"index": 0, "text": "Attention Is All You Need ...", "pageStart": 1, "pageEnd": 1}]
}

Use cases

• RAG pipelines: turn PDFs, Word files and slide decks into Markdown chunks for your vector database
• AI agents: let Claude, ChatGPT or your own agent "read this PDF" or "summarise this PowerPoint" from a link
• n8n, Make and Zapier: extract text from PDF attachments and Office files inside automations
• Document search and archiving: OCR scanned PDFs and faxes into searchable text
• Data extraction: read tables from Excel, PDF reports and invoices as Markdown

How much does it cost?

Pages	Text pages	Scanned pages (with OCR)
100	$0.15	$0.65
1,000	$1.50	$6.50
10,000	$15.00	$65.00

How pages are counted:

• PDF: one page = one PDF page. Pages that need OCR are charged an extra $0.005.
• Images: one image (or TIFF frame) = one page + OCR.
• PowerPoint (PPTX): one slide = one page.
• Excel (XLSX/XLS): one sheet = one page (very large sheets: one page per 50,000 characters of output).
• Word (DOCX), EPUB, HTML, CSV, TXT, JSON: these have no fixed pages, so every 3,000 characters of Markdown output count as one page (a typical Word page).

Failed downloads and documents without readable text are not charged. With the Apify free plan's monthly credit you can convert thousands of pages at no cost.

FAQ

How do I convert a PDF to Markdown? Paste the PDF link into Document URLs and click Start. The Markdown is in the markdown field of the result (also downloadable as JSON, CSV or Excel).

Can it extract text from scanned PDFs (OCR)? Yes. With OCR on auto, pages that have no text layer are rendered and read with an OCR engine (RapidOCR / PaddleOCR models). It works best for printed text in Latin-script languages and Chinese; handwriting is not supported.

Does it keep tables and headings? Yes, as far as the file allows. Excel, Word and PowerPoint tables are converted exactly. In PDFs, tables without borders are detected from the text positions (at least 2 rows and 2 columns that line up) and returned as Markdown tables, with exponents kept as 10^19. PDF headings are recognised from font size, bold text and numbering (1 Introduction, 6.1 Machine Translation) and returned as #/##/### headings. See Known limitations below.

Can I convert Word or PowerPoint files without Microsoft Office? Yes. DOCX, PPTX and XLSX are read directly. Old binary .doc and .ppt files are not supported; save them as .docx/.pptx or PDF first. Legacy .xls works.

What about password-protected or very large files? Password-protected PDFs cannot be opened. Files up to 50 MB are supported; for long documents use pages to convert only what you need.

Is this the same as Microsoft MarkItDown? It uses the open-source MarkItDown library (MIT) for Office, HTML and EPUB files, and adds its own PDF layout analysis (headings, borderless tables), per-page PDF handling, page selection, OCR for scans and images, RAG chunks and a hosted API.

Use it with Claude, ChatGPT and other AI agents

This Actor is made for AI agents: one call turns any document link into Markdown an LLM can read, with page markers it can cite.

1. Add the Apify MCP server to your assistant (e.g. in Claude: Settings → Connectors → add https://mcp.apify.com, then sign in with Apify).
2. Ask in plain language. The assistant finds this Actor, runs it and reads the Markdown.

Example prompts:

• "Read this PDF and give me the key findings with page numbers: https://arxiv.org/pdf/1706.03762"
• "Convert this PowerPoint to text and write a one-page summary for my manager: https://…/deck.pptx"

Tips for agents: use pages (e.g. 1-20) for very long PDFs to get an answer within seconds, and outputChunks when the text goes into a vector store.

You can also call it through the Apify API, on a schedule, or from Make, Zapier and n8n (HTTP or Apify node: pass documentUrls, read markdown).

Pay per call with x402 – no account needed

AI agents can run this Actor and pay per call with x402 (USDC stablecoin payments over HTTP) through Apify's agentic payments – no Apify account or API key required. You pay the same pay-per-event prices listed above. Details: Apify x402 integration.

Known limitations

• PDF tables are rebuilt from text positions, because PDFs have no table structure. Cells that span several columns (e.g. a "BLEU" header over two columns) are put into one of the columns they cover, and cells wrapped over several lines can come out as extra rows.
• Tables inside one column of a two-column page are usually returned as plain text, and tables whose columns do not line up clearly are kept as text rather than turned into a broken table.
• Exponents are recognised from smaller, raised characters (10^19, n^2) inside tables; subscripts are joined to the text (d_model becomes dmodel). In running text, formulas come out as plain text.
• Heading levels follow the numbering (6 → ##, 6.1 → ###) or the font size; documents that use bold body text for headings without numbering may get fewer headings.

Notes

• Only the files you link are downloaded; nothing is crawled, no login or cookies are used. Use documents you are allowed to process.
• Files are processed in memory for your run only and are not shared.
• Built with MIT/Apache/BSD-licensed open-source software: MarkItDown, pdfminer.six, pdfplumber, pypdf, pypdfium2, RapidOCR with PaddleOCR models.

More Actors for AI agents and data pipelines: apify.com/fast_data_asap