PDF Text Extractor: Markdown, Tables & OCR

Give it the addresses of PDF files and get back their content as clean Markdown or plain text, with tables as rows and cells, and the file's metadata. Scanned pages are read with OCR. One row per file, or one row per page if you want to feed the text to an AI or search tool in chunks.

You pay per page that gave text. Pages read from the PDF's own text layer cost less than pages that need OCR. Files that can't be downloaded, aren't PDFs, are password protected or damaged, and blank pages are saved with the reason and cost nothing.

What it does

• Markdown that keeps the structure. Headings come from the document's font sizes (#, ##, ###), wrapped lines are joined into paragraphs, words split by a hyphen at the end of a line are put back together, bullet points become Markdown lists, and tables become Markdown tables in the right place. Two-column pages (papers, reports, newsletters) are read one column at a time.
• Tables you can use. Ruled tables (tables drawn with lines) come back in the tables field as rows of cells with their page number, ready for a spreadsheet.
• OCR only where it's needed. In the default auto mode, a page is read with OCR only when it's a picture with no usable text layer, so a mostly digital PDF with a few scanned pages costs mostly the text-layer price. always reads every page with OCR, for PDFs whose text layer is broken. OCR uses Tesseract with English, German, French, Spanish, Italian, Portuguese and Dutch.
• Metadata. Title, author, subject, creator, producer, creation and modification dates (as ISO dates), page count and file size.
• Page ranges and caps. Read only pages like 1-5,9 or 10-, and at most N pages per file.
• Plain text too. text is the PDF's text layer in reading order, for search indexes or your own parsing. Choose both to get text and Markdown in one run.

Good for: building RAG and AI search indexes from PDF reports and papers, pulling tables out of government and financial reports, archiving forms and filings as text, and checking what a batch of PDFs contains.

How it compares

Most PDF tools in the Store charge either about $1 per 1,000 pages or a flat price per file (about $5 per 1,000 PDFs). Here is what the same work costs:

Job	This Actor	Per-page tools (~$1 / 1,000 pages)	Per-file tools (~$5 / 1,000 PDFs)
1,000 one-page digital PDFs	$1	about $1	about $5
1,000 digital PDFs of 20 pages	$20	about $20	about $5
1,000 scanned PDFs of 20 pages	$80 (OCR)	depends on whether they offer OCR	depends on whether they offer OCR
Files that fail, blank pages	free	check the tool's pricing	check the tool's pricing

So for long digital documents a per-file tool can be cheaper; for short files, mixed batches with failures, and when you want Markdown with tables and OCR that's only paid for where it's used, this one is. Try a few of your own files first: output quality differs most on tables, columns and scans.

Input

Field	What it does
PDF addresses (fileUrls)	Public http(s) addresses of PDF files, one per line. Request-list style objects ({"url": "..."}) work too. Redirects are followed. A PDF sent with the wrong content type is still read; a web page is not.
Output format (outputFormat)	markdown (default), text or both.
Extract tables (extractTables)	Return ruled tables as cells and Markdown tables (default on).
OCR for scanned pages (ocr)	auto (default): only pages with no usable text layer. always: every page. off: never.
OCR languages (ocrLanguages)	Tesseract codes joined with +: eng, deu, fra, spa, ita, por, nld. Default eng.
Pages to read (pageRange)	Optional, for every file: 1-5,9 or 10- (to the end).
Max pages per file (maxPagesPerFile)	Default 500, up to 5,000. Pages past the cap are not read or charged.
One row per page (splitPages)	One row for each page instead of each file.
Files at once (maxConcurrency)	Files downloaded and read in parallel (default 4).

Example input:

{
  "fileUrls": [
    "https://www.rfc-editor.org/rfc/rfc9309.pdf",
    "https://www.irs.gov/pub/irs-pdf/fw9.pdf"
  ],
  "outputFormat": "markdown",
  "extractTables": true,
  "ocr": "auto",
  "pageRange": "1-10"
}

Output

One row per file (values shortened):

{
  "url": "https://files.example.org/quarterly-water-report.pdf",
  "inputUrl": "https://files.example.org/quarterly-water-report.pdf",
  "status": "ok",
  "error": null,
  "note": null,
  "fileName": "quarterly-water-report.pdf",
  "bytes": 3780,
  "pageCount": 3,
  "pagesExtracted": 3,
  "pagesOcr": 0,
  "charCount": 1795,
  "title": "Quarterly Water Report",
  "author": "Harbor City Utilities",
  "subject": "Water quality, first quarter",
  "creator": "Report Builder 2.1",
  "producer": "ReportLab PDF Library",
  "creationDate": "2025-03-14T09:30:00+00:00",
  "modDate": "2025-03-14T09:30:00+00:00",
  "markdown": "<!-- page 1 -->\n\n# Quarterly Water Report\n\n## Summary of findings\n\nHarbor City delivered safe drinking water to every district this quarter...\n\n- Chlorine residual held steady across the north reservoir.\n- Two main breaks were repaired within the target window.\n\n<!-- page 2 -->\n\n## Sampling results by district\n\n| District | Samples | Lead (ppb) | Result |\n|---|---|---|---|\n| North | 48 | 2.1 | Pass |\n...",
  "tables": [
    {"page": 2, "rows": [["District", "Samples", "Lead (ppb)", "Result"], ["North", "48", "2.1", "Pass"], ["Harbor", "52", "3.4", "Pass"]]}
  ],
  "extractedAt": "2026-11-03T18:02:11.000000+00:00"
}

• text and/or markdown hold the pages that gave text, separated by a blank line. In Markdown, each page starts with an invisible <!-- page N --> marker when there is more than one.
• pagesExtracted counts pages that gave text and pagesOcr how many of those were read with OCR. These are the pages you pay for.
• note says when not every page was read and why (page cap, page range, blank pages, scanned pages with OCR off, spending cap).
• With One row per page, each row has page (its number) and source (text for the text layer, ocr, or empty for a blank page) instead of the page counts.

status is one of:

• ok: at least one page gave text. The only status that costs anything.
• no_text: the PDF opened but no page gave text (blank pages, or scanned pages with OCR off).
• download_failed: the server answered with an error, timed out, couldn't be reached, or the file is over 100 MB.
• not_pdf: the address returned a web page or another kind of file.
• encrypted: the PDF needs a password to open.
• invalid_pdf: the file is damaged and can't be read.
• blocked_by_robots: the website's robots.txt asks tools like this one not to download that address.
• invalid_url: not a usable public http(s) address.

Download results as JSON, CSV or Excel, or send them to Google Sheets, Zapier, Make or n8n. The run also saves a SUMMARY record in the key-value store with counts per status, pages extracted, charges and whether the spending cap stopped the run.

Pricing

Pay per event:

• Page extracted: $0.001 ($1 per 1,000 pages) for each page whose text came from the PDF's text layer.
• Page read with OCR: $0.004 ($4 per 1,000 pages) for each scanned page read with OCR.

No start fee, no charge per file, nothing for failed files or blank pages. A 20-page digital report costs $0.02; the same report as a scan costs $0.08. Set a maximum charge per run in the run options: before each file is read, its pages are priced against what's left, a file that would go over is read up to the limit (the row's note says so), and the run stops cleanly.

Good to know

• Files up to 100 MB, downloaded straight to disk. A file stops being read after 25 minutes; the row says how far it got.
• Tables are found where they are drawn with lines. Tables made only of aligned text (no lines) come through as text in reading order, not in tables. Tables on OCR pages aren't detected.
• OCR quality depends on the scan. Clean scans at 200 DPI or more read well; handwriting, very small print and photos of pages read poorly. Pages are rendered at 300 DPI for OCR. Rotated scans should be straightened first.
• Very long documents: if a file's row would be larger than about 8 MB, the full text and Markdown are saved in the run's key-value store, and the row keeps the first million characters with textFileUrl / markdownFileUrl links. Use One row per page to avoid this.
• Password-protected PDFs can't be read. PDFs that only restrict printing or copying (no password to open) are read normally.
• Only public addresses are downloaded: local and internal network addresses are refused, and so are links with a user name or password in them.

Responsible use

• It downloads only the addresses you give it, names itself in its User-Agent (PdfTextExtractorBot), makes one request at a time to each website with at least a second between requests, and respects robots.txt.
• You are responsible for having the right to download and use the documents you process, and for any personal data in them. Follow copyright and the privacy laws that apply to you.

This Actor isn't affiliated with or endorsed by any of the websites it downloads from.

More tools from the same developer

• Image to Text OCR: Read Text from Images & Screenshots
• Website Screenshot & PDF: Full Page, Mobile, Batch URLs