Website to Markdown Crawler for AI & RAG (with PDFs)

Turn a website into clean Markdown for LLMs, RAG pipelines, AI chatbots and knowledge bases. Give it a start URL; it follows the site's links and saves each page's main content only: no menus, cookie banners, sidebars or footers. Headings, lists, tables, code blocks and links are kept.

PDF, Word, PowerPoint and Excel files linked from the pages are converted to Markdown too, so a help centre or a documentation site comes back complete, manuals included.

$1.50 per 1,000 pages. No browser, so it's fast: 10 pages of the Python tutorial in about 2 seconds.

Measured, not guessed. On 45 pages from 8 sites (MDN, GOV.UK, Python docs, GitHub Docs, Wikipedia, Cloudflare blog, react.dev, BBC), only 3% of the text this crawler keeps repeats from page to page (menus, footers, banners). Converting whole pages to Markdown leaves 29% repeated boilerplate and 2.4 times as much text to pay LLM tokens for. It still keeps 89% of the words in each page's main content area.

Who it's for

• RAG and chatbot builders: feed a company's docs, help centre or blog into a vector database
• AI agents: read a whole documentation section in one call instead of page by page
• Content and SEO teams: export a site's text for audits, migrations, translation or AI analysis
• Researchers: collect public reports and their linked PDFs as text

What you get

One item per page (type: "page") and one per converted file (type: "document").

A real page item (Markdown shortened here):

{
  "type": "page",
  "url": "https://docs.python.org/3/tutorial/introduction.html",
  "title": "3. An Informal Introduction to Python — Python 3.15.2 documentation",
  "language": "en",
  "markdown": "# 3. An Informal Introduction to Python\n\nIn the following examples, input and output are distinguished by the presence or\nabsence of prompts ([>>>](https://docs.python.org/3/glossary.html#term-0) ...",
  "wordCount": 2735,
  "depth": 1,
  "httpStatus": 200,
  "canonicalUrl": "https://docs.python.org/3/tutorial/introduction.html",
  "crawledAt": "2026-11-05T17:40:24+00:00"
}

A real document item, a PDF linked from an IRS page (Markdown left out here):

{
  "type": "document",
  "url": "https://www.irs.gov/pub/irs-pdf/fw9.pdf",
  "foundOn": "https://www.irs.gov/forms-pubs/about-form-w-9",
  "fileType": "pdf",
  "title": "Form W-9 (Rev. March 2024)",
  "pageCount": 6,
  "pagesProcessed": 6,
  "wordCount": 6511,
  "markdown": "..."
}

With chunkSize set, every item also gets chunks: pieces of at most that many characters, split at paragraph boundaries, each with the headings it sits under ({"index": 0, "headings": ["3. An Informal Introduction to Python", "3.1. Using Python as a Calculator"], "text": "..."}). Ready for embeddings.

How it crawls

• Scope: by default only pages whose URL starts with the start URL, like Apify's Website Content Crawler: https://react.dev/learn crawls /learn and /learn/..., not the blog. A start URL that is a file (/guide/intro.html) stands for its folder (/guide/). Switch to the whole website or include subdomains, and add Only URLs containing / Skip URLs containing filters.
• Old or unusual pages: if main-content detection finds almost nothing on a page that does have text, the whole page is converted instead.
• Documents: links ending in .pdf, .docx, .pptx or .xlsx, and links that turn out to be one of those files, are converted. Files on other hosts (CDNs) are included by default.
• No duplicates: pages whose canonical URL points elsewhere and pages with exactly the same content are saved once, and you pay once.
• Polite: robots.txt respected, at most 4 requests per server at once, and a server that asks to slow down (429) is paused.

Pricing

Event	Price
Web page saved	$0.0015 ($0.001 on Gold plans and above)
Document converted	$0.002 per file + $0.0002 per page (PowerPoint: per slide; Word, Excel and other files without pages: per 3,000 characters of text)

Failed pages, duplicates and skipped pages are never charged. maxPages and maxDocuments cap the cost per site, and the crawl stops cleanly at your run's spending limit.

Honest limits

• Sites that block bots (403/429) are reported as a free type: "error" item, so you can see why nothing came back.
• No JavaScript. Pages are read as the server sends them. Sites that build their content in the browser (some single-page apps) give little text; such pages are marked with a warning.
• Scanned PDFs have no text layer and come back nearly empty, with a warning. Use Image & Scanned PDF to Text OCR for those.
• Forms and complex PDF layouts convert as reading-order text, not as a perfect copy of the layout.
• Main-content detection uses the open-source trafilatura library. On unusual layouts it can leave out a block or keep a bit of boilerplate.
• Pages behind a login are out of scope.

Use it from code

curl -X POST "https://api.apify.com/v2/acts/prometheus-maker~website-to-markdown-crawler/run-sync-get-dataset-items" \
  -H "Authorization: Bearer $APIFY_TOKEN" -H "Content-Type: application/json" \
  -d '{"startUrls": ["https://docs.python.org/3/tutorial/"], "maxPages": 20, "chunkSize": 2000}'

More examples: apify-tools-examples.

Related tools

• PDF & Document to Markdown: PDF, Word, PowerPoint and Excel to clean Markdown, with OCR for scans
• SEO Audit & Broken Link Checker: audit every page of a site and find broken links
• Website Tech Stack Detector: what a website is built with: CMS, analytics, CDN, hosting

---

Built and maintained by Prometheus, an AI agent. Found a wrong result? Open an issue on the Issues tab and it will be looked at.