Skip to main content
PPDF123

OCR a Scanned PDF

OCR PDF reads a scanned or image-based PDF and returns either a searchable PDF with a hidden text layer or its text as a Markdown file. The original file is unchanged.

Drag and drop files here, or click to choose

Accepts PDF · up to 500.0 MB per file · up to 20 files at once

You can also paste a file with Ctrl/Cmd+V, or from the clipboard menu.

OCR PDF reads a scanned or image-based PDF and returns it in one of two forms. Searchable PDF gives you the same pages with an invisible text layer, so you can search, select and copy the text. Markdown gives you the text as a .md file for notes, repositories and pipelines. The original file is never changed.

A searchable PDF is made with Tesseract. Every page is rendered at 300 dpi, recognized in the language you choose, and written back as a page image with the recognized text hidden behind it. The page count stays the same. Because every page is re-rendered, a page that already had real text becomes an image with a text layer, and the file size can differ a lot from the original. Pick one of 12 languages in the form; the API also accepts up to four joined by a plus sign, such as eng+fra. At most 50 pages are processed in one request.

Markdown output uses a different engine, which fuses native PDF text with optical recognition on image pages. Force OCR (`ocrType=force-ocr`, the portal default) re-recognizes every page, while Normal uses existing text where the file has it and recognizes image-only pages. This engine has no language setting, and the language list does not apply to it.

Accuracy follows scan quality, contrast and skew in both modes. Crop and straighten phone captures before upload when you can, because a trapezoid photo of a page wastes recognition on background. OCR recognizes characters; it does not translate or correct the text, and it does not clean or deskew the scan itself.

Handwriting, decorative typefaces and low-resolution faxes are the hardest inputs. Where a character is ambiguous, such as a 1 against an l or a 0 against an O, the engine picks the most likely reading and moves on, and neither mode returns a confidence score. Spot-check names and numbers that must be exact. In Markdown, tables become pipes and a merged cell can lose its row, and multi-column pages are read in the order the engine infers.

Do not OCR documents you are not allowed to digitize or redistribute. Jobs run server-side, so download the result promptly. We do not keep an OCR library of your files. Password-protected files are unlocked for you on this page once you enter the password; API callers run Unlock first.

Features

  • Searchable PDF with a hidden text layer, or Markdown text
  • Searchable PDF: Tesseract at 300 dpi in 12 languages, up to 50 pages
  • Markdown: native text plus recognition on image pages; Force or Normal mode
  • The original PDF is unchanged
  • Anonymous API at /api/v1/misc/ocr-pdf with outputFormat and languages

When to use this tool

  • Make a stack of scanned contracts searchable in a document archive
  • Copy a paragraph out of a photographed or scanned page
  • Feed scans to agents or pipelines that want Markdown
  • Make photographed receipts searchable in a notes app

How do I OCR a scanned PDF?

  1. Upload a scanned or camera-captured PDF.
  2. Choose Searchable PDF or Markdown. For a searchable PDF, pick the language of the text.
  3. Click Process and download the result.
  4. Spot-check names and numbers before you rely on it, and try searching the PDF for a word you know is in it.

Limits and edge cases

  • Accuracy varies with image quality; handwriting and stylized fonts recognize poorly
  • Searchable PDF: at most 50 pages, 300 dpi, one image per page
  • Markdown has no language setting, confidence scores or per-page output
  • Not translation, deskewing or cleaning of the scan
  • Very large page counts take longer
  • Upload limit on this website: 500 MB per file, sent in chunks above about 95 MB. A single direct API request body is capped at 100 MB.

Examples

  • A 20-page grayscale contract scan becomes a PDF in which a search for a clause number finds the page
  • A scanned page with a line of English text returns a PDF from which that line can be copied
  • A skewed phone photo may need re-capture before OCR is useful

Privacy for this tool

Files are uploaded over HTTPS, processed in memory or a short-lived temporary directory on our servers, and deleted when your result is ready. We do not keep copies for later browsing, training, or advertising profiles. See the Privacy Policy for retention details and AdSense cookie disclosures.

Frequently asked questions

Can I get a searchable PDF?
Yes. Choose Searchable PDF in the form. Every page is recognized and written back as an image with a hidden text layer, so you can search and copy its text.
Why does the searchable PDF look like a scan, not like my original file?
Every page is rendered to an image before recognition, including pages that already had real text. The text layer sits behind the image.
Which languages are supported?
The form offers English, German, French, Spanish, Portuguese, Italian, Russian, Arabic, Simplified Chinese, Traditional Chinese, Japanese and Korean. Through the API, up to four codes can be joined with a plus sign, such as eng+fra.
How many pages can I process?
Up to 50 pages for a searchable PDF. A longer file is rejected with an error; split it first.
Does the Markdown output use the language setting?
No. The Markdown engine has no language parameter. The language list is only for searchable PDFs.
Will OCR change how pages look?
No. You get a new file; the original is unchanged. A searchable PDF shows each page as an image, as a scan does.
Does OCR handle handwriting?
Poorly. Both engines target machine-printed characters. Cursive, decorative fonts and low-DPI faxes often come back as noise.
Do I need Force OCR on a born-digital PDF?
Usually no. For Markdown, Normal uses the existing text and skips the recognition engine. Use Force if that layer is broken.

Last updated:

Call this from code

Every tool on this site is a plain REST endpoint - no account or API key needed for anonymous use. Built for AI agents and developers as much as for browsers.

curl -X POST "https://pdf123.xyz/api/v1/misc/ocr-pdf" \
  -F "[email protected]" \
  -F "outputFormat=markdown" \
  -F "ocrType=force-ocr" \
  -F "languages=eng" \
  -o output.pdf

Also available as an MCP tool for agent clients that speak Model Context Protocol (JSON-RPC 2.0 over POST /mcp). Full API reference

Use it from an AI agent

Skill

Claude Code, Codex, Cursor and other AI agents can run OCR PDF for you with this skill: /skills/pdf123-ocr.md

All agent skills

Files are used only for this processing job and deleted automatically afterward.