PDF Text Extractor - Page by Page, With OCR for Scans
Pricing
from $3.00 / 1,000 page extracteds
PDF Text Extractor - Page by Page, With OCR for Scans
Extract text from any list of PDF URLs, one row per page, with title, author and page count. Scanned pages with no text layer are read by OCR in the same run, and each row says how it was read. 20 languages. Pay only for pages with text.
Pricing
from $3.00 / 1,000 page extracteds
Rating
0.0
(0)
Developer
Kobkoon Janngam
Maintained by CommunityActor stats
0
Bookmarked
1
Total users
0
Monthly active users
2 hours ago
Last modified
Categories
Share
PDF Text Extractor — page-by-page text, with OCR for scanned pages
Get the text out of any list of PDFs, one row per page. Each row also carries the document's title, author and page count. A scanned page has no text layer, so it is read by OCR in the same run, and each row says how its page was read. Built for data pipelines, RAG and AI agents.
Quick start: press Start with no configuration. It reads three sample PDFs: a 14-page research paper with a text layer, and two scans that need OCR. Then paste your own links.
Sample output
| fileName | pageNumber | pageCount | method | ocrConfidence | text |
|---|---|---|---|---|---|
| compressed.tracemonkey-pldi-09.pdf | 1 | 14 | text-layer | — | Trace-based Just-in-Time Type Specialization for Dynamic Languages… |
| compressed.tracemonkey-pldi-09.pdf | 2 | 14 | text-layer | — | Hence, recording and compiling a trace speculates that… |
| linn.pdf | 1 | 1 | ocr | 91 | The LinnSequencer · 32 Track MIDI Sequence Recorder… |
| skew.pdf | 1 | 1 | ocr | 91 | The LinnSequencer is a state-of-the-art composition and performance tool… |
These rows come from the default run on 30 October 2026. It read 16 pages with text, 2 of them by OCR.
Every row also carries pdfUrl, status, charCount, wordCount, title, author, createdAt and chunk/chunkCount, plus a note when something needs explaining.
What you can do with it
- Build a RAG index. Rows arrive already split into chunks under 3 KB, broken at paragraphs, and each one keeps its page number.
- Give an AI agent a PDF. It gets one page at a time instead of a file it cannot open.
- Digitise scanned archives. Contracts, invoices, old manuals and faxes become searchable text.
- Monitor reports. Pull the text of filings, papers and policy documents for analysis.
Input
| Field | Default | Notes |
|---|---|---|
pdfUrls | 3 samples | Up to 100 MB each and 200 per run. Each link must open without signing in. |
maxPagesPerPdf | 50 | Reads at most this many pages of each PDF, starting from page 1. Up to 2,000. |
ocrScannedPages | true | Reads pages that have no text layer with OCR. Turn it off for faster runs on digital-only PDFs. |
languages | eng | 20 OCR languages. A text layer is read in any language. |
maxTextBytesPerRow | 3000 | Splits long pages across rows. 0 returns one row per page. |
Every page gets a row that says what happened
| status | method | Meaning | Charged |
|---|---|---|---|
ok | text-layer | Text read straight from the PDF | Yes, once per page |
ok | ocr | A scanned page read by OCR; see ocrConfidence | Yes, once per page |
empty | none / ocr | The page is blank, or OCR was turned off for a scan | No |
download-failed / not-a-pdf / unreadable | — | One row per file, with the reason. Covers dead links, web pages, broken or password-protected files | No |
One bad file never fails the run.
How it works
Pages are read with pdf.js, the PDF engine inside Firefox. When a page has no text, it is rendered and read with Tesseract OCR. Both engines run inside the Actor, so there is no API key and no third-party service. JBIG2, CCITT and JPEG 2000 scans are all decoded.
A text layer is reproduced exactly as the PDF stores it. Charts and diagrams sometimes contain stray glyphs, and those come through too. OCR quality depends on the scan, and ocrConfidence tells you which pages to trust.
Pricing
You pay per page returned with text, whether it was read from the text layer or by OCR. Blank pages, pages beyond your limit and files that fail cost nothing.
Works well with
Same publisher, same output conventions, same honest-failure rules:
- image-to-text-ocr — The text in any list of images: screenshots, scans, receipts. 20 languages.
Disclaimer
This Actor reads only the files you link to: it scrapes no website and calls no third-party API — the OCR engine runs inside the Actor. Make sure you have the right to process the files you submit.
