PDF & Document to Markdown: OCR, Tables, Word, Excel

Turn any document into clean Markdown that LLMs and RAG pipelines can actually use.

Give this Actor links to your files, or upload them, and get back structured Markdown: headings, paragraphs, lists and real tables, not a wall of text. It reads PDFs (including scanned PDFs, with OCR in 27 languages), Word, PowerPoint, Excel, CSV, HTML, EPUB, plain text and images. Choose one row per document, one row per page, or ready-to-embed chunks that keep their page numbers and section headings.

✨ Why use it

• 📑 Tables stay tables: ruled tables and borderless (academic, booktabs-style) tables come out as Markdown tables, and also as rows of cells you can load straight into a spreadsheet or a database.
• 🧭 Structure, not just text: headings from font sizes and bold section titles, bullet and numbered lists, multi-column reading order, hyphenated words joined, running headers, footers and page numbers removed.
• 🔍 OCR only where it is needed: pages without a usable text layer (scans, photos, broken fonts) are read with OCR automatically, and you pay the OCR price only for those pages.
• 🗂️ Nine formats, one output: PDF, DOCX, PPTX (with speaker notes), XLSX/XLS, CSV, HTML, EPUB, TXT/MD and PNG/JPG/TIFF/WebP images all come out in the same shape.
• 🧩 RAG-ready chunks: paragraph-aligned chunks of the size you choose, with overlap, pageStart/pageEnd for citations and the heading path (["4 Policies", "4.2 Expenses"]) for context.
• 🛡️ Robust on bad files: each document is converted in its own process with a timeout. A broken or huge file becomes a failed row with a clear message, is not billed, and does not stop the run.
• 💸 Pay per page: no subscription, and OCR from $3 per 1,000 pages.

🗂️ Supported files

Type	What you get
PDF (text)	Headings, paragraphs, lists, tables, metadata (title, author, dates). Password-protected PDFs work with PDF password.
PDF (scanned)	OCR text per page, in the languages you pick.
Images	PNG, JPG, TIFF (every page of a multi-page TIFF), WebP, BMP, GIF, read with OCR.
Word .docx	Headings from the document's styles, bold and italic, lists, links and tables.
PowerPoint .pptx	One section per slide, with its title, bullets with their levels, tables and speaker notes.
Excel .xlsx .xls, CSV	One Markdown table per visible sheet, with dates and numbers as you see them. The CSV delimiter and encoding are detected.
HTML, EPUB	Clean Markdown without scripts, menus or forms. EPUB chapters in reading order.
Text .txt .md	As is, tidied.

The type is detected from the file's content, so links without a file extension work too. Old binary .doc and .ppt, OpenDocument and RTF files are not supported yet: save them as .docx, .pptx or .pdf.

💡 Use cases

• RAG and AI assistants: load manuals, policies, contracts, reports and research papers into a vector database, with page numbers to cite.
• LLM pipelines: give a model clean Markdown instead of raw PDF text, so it sees headings and tables and uses fewer tokens.
• Data extraction: pull the tables out of annual reports, price lists, statistics releases and invoices into rows and columns.
• Archives and back offices: make scanned letters, forms and old reports searchable with OCR.
• Automation: convert the attachments that arrive by email or in a cloud folder, through Make, Zapier, n8n or the Apify API.

🚀 How to use it

1. Click Try for free.
2. Paste the links to your documents in Document URLs, or upload files.
3. Pick the Output format: one row per document, one row per page, or RAG chunks. For scans in another language, add it under OCR languages.
4. Run it and download the result as JSON, CSV or Excel, or read it from the Apify API. With Save .md files on, each document is also saved as a .md file.

📥 Input examples

Two PDFs and a Word file, one row per document:

{ "urls": ["https://www.irs.gov/pub/irs-pdf/fw9.pdf", "https://www.irs.gov/pub/irs-pdf/p15.pdf", "https://example.com/policy.docx"] }

Chunks of about 500 tokens for a vector database, with page markers:

{ "urls": ["https://example.com/handbook.pdf"], "outputFormat": "chunks", "chunkSize": 2000, "chunkOverlap": 200 }

Scanned Portuguese and English letters, every page read with OCR:

{ "urls": ["https://example.com/scan.pdf"], "ocr": "always", "ocrLanguages": ["por", "eng"] }

From code, send files inline as base64:

{ "base64Documents": [{ "fileName": "invoice.pdf", "data": "JVBERi0xLjcK..." }] }

📦 Output

One row per document (default):

{
  "url": "https://www.irs.gov/pub/irs-pdf/fw9.pdf",
  "fileName": "fw9.pdf",
  "fileType": "pdf",
  "status": "ok",
  "title": "Form W-9 (Rev. March 2024)",
  "pageCount": 6,
  "pagesConverted": 6,
  "ocrPages": 0,
  "tableCount": 1,
  "words": 6464,
  "markdown": "…\n\n### Line 3a\n\nCheck the appropriate box on line 3a for the U.S. federal tax classification…\n\n| IF the entity/individual on line 1 is a(n) . . . | THEN check the box for . . . |\n|---|---|\n| • Corporation | Corporation. |\n| • Partnership | Partnership. |\n…",
  "tables": [{ "page": 3, "rows": [["IF the entity/individual on line 1 is a(n) . . .", "THEN check the box for . . ."], ["• Corporation", "Corporation."]] }],
  "convertedAt": "2026-10-18T22:12:27Z"
}

With RAG chunks, each row is one chunk:

{
  "url": "https://example.com/handbook.pdf",
  "fileName": "handbook.pdf",
  "chunkIndex": 12,
  "chunkCount": 87,
  "text": "## 4.2 Expenses\n\nEmployees can claim travel expenses when…",
  "pageStart": 9,
  "pageEnd": 10,
  "headings": ["4 Policies", "4.2 Expenses"],
  "characters": 1984
}

Other fields: author, createdAt, modifiedAt, characters, text (with Plain text), pages (with Per-page Markdown), markdownFileUrl (with Save .md files) and message, which says why a document failed or was cut by a limit. The dataset has three views: Documents, Pages and RAG chunks.

💰 How much does it cost?

Event	Price
Document converted	$0.002
Page converted (text layer)	$0.0003
Page read with OCR	$0.003

Word, PowerPoint, Excel, CSV, HTML, EPUB and text files count one page for every 3,000 characters of Markdown they produce. Higher Apify plans get volume discounts.

Examples: a 10-page PDF costs $0.005. A 200-page report costs $0.062. A 20-page scanned contract costs $0.062. 1,000 single-page invoices cost $2.30. Files that fail are not billed. You can set a maximum cost per run in the run options: the Actor stops before a document it cannot pay for, and never bills a document it does not deliver.

🔍 Good to know

• Scanned PDFs: OCR on Auto reads only pages without usable text. Use Always when a PDF has a text layer that is wrong (garbled characters, a bad OCR from another tool). Max OCR pages per document (default 200) protects you from surprise costs on very long scans.
• Tables: tables with ruled lines and booktabs-style tables (horizontal rules only) are detected. Tables laid out with spaces alone, with no lines at all, come out as text. In Word and HTML tables, merged cells are repeated in each column they span.
• Very long documents: the Markdown field holds up to 1.5 million characters. Longer documents are saved whole as a .md file, linked in markdownFileUrl.
• Limits: files up to 500 MB (Max file size, default 100 MB) and up to 10,000 pages per document (Max pages per document, default 1,000).
• Downloads: links must be directly downloadable without a login. Files you upload are read from your own key-value store.
• Privacy: your files are processed only inside your own Apify run and are not kept by this Actor after the run. The results are stored in your run's dataset, under your account's data retention settings.

❓ FAQ

Does it use an AI model or send my files anywhere? No. Conversion and OCR run inside the Actor with open-source engines (pdfplumber, pdfium, Tesseract). Nothing is sent to an external AI service.

How good is the OCR? It is Tesseract 5 at 300 dpi, which reads clean printed text very well. Handwriting and low-quality photos give weaker results.

Can I get one row per page? Yes: set Output format to One row per page. Slides, sheets and EPUB chapters count as pages.

Can I convert web pages? You can pass an HTML link, and the Actor converts that one page. To crawl a whole website, use a crawler and send it the pages.

Something looks wrong? Open an issue on the Actor's Issues tab with a link to the file (or a similar one) and the input you used.

🔗 More from the same developer

• European Public Data MCP Server: official European data as tools for your AI agent.
• Portuguese Case Law: Portuguese court decisions with full text, for legal AI.
• EU Company Verification (KYB): bulk checks in 9 official registries.
• Public Tenders: Portugal, Spain & EU: a daily feed of new calls for tender.