PDF to Markdown & JSON Extractor — LLM-Ready

Turn any PDF into clean, LLM-ready Markdown and structured JSON. Point this PDF to JSON / PDF to Markdown converter at one or many PDF URLs and get back tidy Markdown (text plus tables), plain text, and document metadata — one structured record per PDF. Built for feeding LLMs, RAG pipelines, AI agents, and fine-tuning datasets clean document content instead of raw, messy PDF bytes.

No headless browser, no external services — just fast, reliable extraction.

What it does

• PDF to Markdown — converts a PDF's text into clean Markdown ready to drop into an LLM prompt or vector database.
• Table extraction — detects tables and renders them as Markdown pipe tables, so structure survives the conversion.
• Metadata — title, author, subject, keywords, producer, and creation/modification dates from the PDF's own info dictionary.
• Per-document or per-page output — one record per PDF by default, or one record per page (splitPages) for easy RAG chunking.
• Page ranges & caps — extract only the pages you need (pageRange) and bound cost with maxPagesPerPdf.
• Structured output — exportable to JSON, JSONL, CSV, or Excel, or via the Apify API.

Use cases

• Build a RAG knowledge base from reports, papers, manuals, or contracts.
• Feed LLM agents clean document text instead of raw PDF.
• Assemble fine-tuning / training datasets from public PDFs.
• Convert research papers, invoices, or datasheets into structured, queryable data.

Input

Field	Description
startUrls	Direct links to the PDF files to extract.
outputFormat	markdown (LLM-ready) or text.
extractTables	Detect tables and render them as Markdown pipe tables.
splitPages	Output one item per page instead of one per document.
pageRange	Pages to extract, e.g. 1-10 or 3 (empty = all).
maxPagesPerPdf	Hard cap on pages parsed per document (main cost lever).
maxFileSizeMb	Skip PDFs larger than this, without charging.

Output

Each PDF becomes one dataset item (or one per page with splitPages):

{
  "url": "https://arxiv.org/pdf/1706.03762",
  "title": "Attention Is All You Need",
  "content": "## Page 1\n\nClean markdown of the page text...\n\n| Layer | Complexity |\n| --- | --- |\n| Self-Attention | O(n²·d) |",
  "format": "markdown",
  "wordCount": 8123,
  "pageCount": 15,
  "totalPages": 15,
  "metadata": {
    "author": "Vaswani et al.",
    "creationDate": "D:20170606",
    "producer": "pdfTeX",
    "subject": null,
    "keywords": null
  }
}

Pricing

Pay-per-result: you're charged per page successfully parsed — no monthly fee, and no charge for PDFs that fail, are password-protected, are too large, or have no extractable text.

Notes

• Works on text-based PDFs (papers, reports, docs, invoices, datasheets). Scanned / image-only PDFs have no text layer; they're detected and skipped free. OCR is on the roadmap (future toggle).
• Password-protected and corrupt PDFs are reported as error records (not charged).
• Processes public PDF URLs; it does not log in or bypass access controls.

Licensing

Built entirely on permissively licensed libraries — pdfplumber (MIT), pdfminer.six (MIT), pypdfium2 (BSD/Apache), and Pillow. No AGPL components.

FAQ

Does it keep tables? Yes — with extractTables on and Markdown output, detected tables are rendered as Markdown pipe tables beneath the page text.

Can I extract just some pages? Yes — set pageRange (e.g. 1-5) and/or maxPagesPerPdf.

What about scanned PDFs? They have no text layer, so v1 skips them for free (no charge). OCR support is planned.

What formats can I export? JSON, JSONL, CSV, or Excel, or via the Apify API.