Crawlsmith

Document to Markdown

Convert PDF, Word, PowerPoint, Excel, CSV, HTML, EPUB and images from URLs to clean Markdown or text: headings, lists and tables kept, OCR for scans, RAG-ready chunks.

What it does

This tool downloads every document you list and converts it into structured, ready-to-use text. It rebuilds the document's outline instead of dumping raw characters: titles become # headings, numbered sections like 3.1 Encoder become sub-headings, bullet points become lists, Word and Excel tables become Markdown tables, and running headers, footers and page numbers are removed. No manual copy-pasting, no broken line wraps.

Two ways to get your results:

  1. One row per document: the full Markdown and/or text of each file, with title, author, dates, page and word counts.
  2. One row per chunk: documents split into token-sized pieces with their section path, ready to embed and load into Pinecone, Weaviate, Qdrant, pgvector or any vector database.

Who uses it

Use Case How It Helps
AI & RAG developers Feed clean, chunked Markdown into embeddings and chatbots instead of garbled PDF text
Legal & compliance teams Convert contracts, policies and filings into searchable text with their structure intact
Researchers & analysts Extract papers and reports into Markdown for notes, summaries and literature reviews
Back-office automation OCR scanned invoices, forms and receipts into text for downstream parsing
Knowledge-base builders Migrate Word, PowerPoint and Excel files into wikis, Notion or static sites
No-code builders Plug document text into Make, Zapier or n8n flows with one API call

Input

15 options. Required ones are marked.

Option What it does
Document URLs required Links to PDF, Word (.docx), PowerPoint (.pptx), Excel (.xlsx/.xls), CSV, HTML, EPUB or image files.
Output format Markdown keeps headings, lists, tables and links.
One row per One dataset row per document, or one row per chunk (ready for a vector database).
Chunk size (tokens) Approximate tokens per chunk.
Chunk overlap (tokens) Tokens repeated from the end of the previous chunk.
OCR Auto reads scanned pages and images only.
OCR languages Languages printed in your scans.
Max pages per document 0 = all pages (PDF).
Max rows per sheet Spreadsheets and CSV.
Include per-page Markdown Adds a pages array (page number + Markdown) to each document row.
Include PowerPoint speaker notes Adds the presenter notes under each slide.
PDF password For password-protected PDFs.
Max file size (MB) Larger files are skipped with an error row.
Files in parallel How many documents are converted at the same time.

Plus 1 more options. Full input reference on Apify.

Output fields

12 fields, grouped the way the dataset views show them. Export to JSON, CSV, Excel or pull them through the API.

Documents

  • URL
  • Type
  • Title
  • Pages
  • OCR pages
  • Words
  • Error

Content

  • Title
  • Markdown
  • Text

Chunks

  • URL
  • #
  • Section
  • Tokens
  • Text

Example output

One record, shortened. Real runs return one row like this per result.

{
  "url": "https://calibre-ebook.com/downloads/demos/demo.docx",
  "fileName": "demo.docx",
  "fileType": "docx",
  "title": "DOCX Demo",
  "author": "Kovid Goyal",
  "created": "2013-06-05",
  "modified": "2013-06-20",
  "pageCount": 1,
  "pagesProcessed": 1,
  "ocrPages": 0,
  "sizeBytes": 1311881,
  "wordCount": 1648,
  "markdown": "# Demonstration of DOCX support in calibre\n\nThis document demonstrates the ability of the calibre DOCX Input plugin...\n\n# Text Formatting\n\n## Inline formatti...",
  "scrapedAt": "2026-11-16T12:00:00.000Z"
}

How to run it

  1. 1. Open the scraper on Apify and sign in (free plan works).
  2. 2. Fill in the input: search terms, URLs or filters, and a result limit.
  3. 3. Run it, then download JSON, CSV or Excel, or pull the dataset through the API.
  4. 4. Schedule it to get fresh data daily or weekly.

FAQ

Does it work with scanned PDFs? +

Yes. In Auto mode every page without a text layer is read with OCR. Choose the languages printed in your documents for the best accuracy.

Are tables preserved? +

Tables from Word, Excel, PowerPoint and web pages become Markdown tables. Tables inside PDFs come out as text in reading order, because PDFs do not store table structure.

Can I convert password-protected or private files? +

Protected PDFs work with the PDF password option. Files must be reachable by a link: public URLs, signed links (S3, Google Cloud Storage, Dropbox) or your own server.

What chunk size should I use for RAG? +

300 to 800 tokens with 50 to 100 tokens of overlap is a good default for most embedding models.

Is there a size limit? +

50 MB per file by default, adjustable up to 500 MB.

Need something this scraper does not do?

Extra fields, a different filter, a recurring feed into your database or a one-off bulk export.

Ask for a custom version

Related scrapers