PDF Text Extractor

Turn PDF links into plain text for search, RAG or AI agents. Each PDF is read page by page with Mozilla's PDF.js 1: text with pages separated, page and word counts, and a needsOcr flag for scanned pages. No OCR, no browser.

What it does

• Text layer only, page by page; each page is released before the next, so long documents don't grow memory.
• Scans flagged, not guessed: a page with no text layer is marked needsOcr and listed in pagesNeedingOcr. Usually a scan or outlined text; blank pages are flagged too. No OCR is run or charged.
• Pages are separated by a form feed (\f) in text, as pdftotext does. pages gives characters, words and needsOcr for each page.
• Checks:
  • the file must start with the %PDF- header 2;
  • password-protected or damaged files give a clear error;
  • redirects are followed (up to 5).
• Limits you set: file size, pages, characters and time per PDF. Pages read before a limit are kept; stoppedEarly says why.
• Safety: only http and https. Private, loopback, link-local and cloud metadata addresses are refused 3, including after DNS and on every redirect.

Quick start

The prefilled input reads one GOV.UK PDF:

{
  "urls": [
    "https://assets.publishing.service.gov.uk/media/5a7e3de4ed915d74e622507f/legal-aid-checklist.pdf"
  ]
}

The row from the run on 4 October 2026, without text and pages:

{
  "index": 0,
  "url": "https://assets.publishing.service.gov.uk/media/5a7e3de4ed915d74e622507f/legal-aid-checklist.pdf",
  "finalUrl": "https://assets.publishing.service.gov.uk/media/5a7e3de4ed915d74e622507f/legal-aid-checklist.pdf",
  "status": 200,
  "title": "Civil representation - Checklist for new applications - August 2013",
  "pageCount": 4,
  "pagesRead": 4,
  "chars": 2531,
  "words": 492,
  "needsOcr": false,
  "pagesNeedingOcr": [],
  "textTruncated": false,
  "stoppedEarly": null,
  "bytes": 157341,
  "error": null
}

text starts "Civil Representation\nChecklist for new applications\nAugust 2013 v2".

Input

Field	Default	Meaning
urls	(required)	PDF links, up to 1,000
maxPages	500	Pages read per PDF
maxFileMB	30	Largest file downloaded
maxCharsPerPdf	1,000,000	Text kept per PDF
timeoutSecs	60	Download and reading, per PDF
maxConcurrency	2	Most PDFs at once (1–4), within memory

Output

One row per PDF, with index giving its position in urls. Rows arrive as each PDF finishes.

Field	Meaning
text	Page texts joined by \f
pages	[{page, chars, words, needsOcr}]
needsOcr, pagesNeedingOcr	Scanned pages, if any
title, pageCount, pagesRead, chars, words	From the document
status, finalUrl, bytes	The download
textTruncated, stoppedEarly, error	What stopped it

Use from code

curl -X POST "https://api.apify.com/v2/acts/landlordtools~pdf-text-extractor/run-sync-get-dataset-items?token=YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"urls": ["https://assets.publishing.service.gov.uk/media/5a7e3de4ed915d74e622507f/legal-aid-checklist.pdf"]}'

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: 'YOUR_API_TOKEN' });
const run = await client.actor('landlordtools/pdf-text-extractor').call({ urls: ['https://example.org/report.pdf'] });
const { items } = await client.dataset(run.defaultDatasetId).listItems();
console.log(items[0].text.split('\f')[0]);

Use with AI agents

Agents can call it through Apify's MCP server (https://mcp.apify.com) to read a PDF instead of guessing what it says.

Accuracy

Tested on 4 and 10 October 2026 against poppler's pdftotext on 200 GOV.UK PDFs (7,785 pages), using word-level F1:

• median 0.998 per PDF;
• 0.988 over all words pooled;
• 191 of the 200 PDFs at 0.95 or above.

The lowest scores are forms and multi-column layouts. No errors on either side. Details: tests/benchmark/.

Limits

• No OCR: scanned pages come back empty and flagged needsOcr. A scan with a few typed characters on it (such as a page number) isn't flagged.
• Memory: 256 MB reads one PDF at a time; 512 MB reads two.
• Reading order follows the PDF's text positions. Multi-column layouts and tables may interleave.
• Forms: field values typed into a form aren't part of the text layer.

Integrations

Power Platform, OpenAPI, Excel and Google Sheets files are in integrations/README.md. Apify's Zapier, Make and n8n apps work too.

Pricing

Pay per event: $2 per 1,000 results ($0.002 each), plus Apify's minimal start fee. A result is one PDF read, whatever its length. PDFs that can't be fetched or read are free, and the run stops at your spending limit 5.

Related Actors

• Link and Citation Checker: check that links and quotes are right.
• Sitemap URL Extractor: find every page on a site.

Sources

1. pdfjs-dist (Mozilla PDF.js), Apache-2.0. Accessed 4 October 2026.
2. Library of Congress: PDF 1.7 (ISO 32000-1:2008) format description, file header %PDF-. Accessed 4 October 2026.
3. IANA IPv4 Special-Purpose Address Registry. Accessed 4 October 2026.
4. IANA IPv6 Special-Purpose Address Registry. Accessed 4 October 2026.
5. Apify: pay-per-event pricing. Accessed 4 October 2026.