OCR API (Image to Text, Searchable PDF, 20 Languages)

Turn scans, phone photos, screenshots and scanned PDFs into text, one row per page, as plain text, Markdown, word boxes with confidence or a searchable PDF, in 20 languages. Pages of a PDF that already have selectable text are read straight from the file, for free. No login, no API key.

• Make scanned archives searchable: add an invisible text layer to old scans so any PDF viewer can search and copy them.
• Read receipts and labels: narrow columns are read line by line, so prices stay on the line of their item.
• Feed documents to an LLM or RAG pipeline: get clean text or Markdown from scans that text extractors return empty.
• Build redaction and form tools: every line and word comes with a pixel box and a 0-100 confidence.
• Process mixed batches: images and PDFs in one run, and only the pages that need OCR are charged.

Quick start

Paste this into the JSON tab of the input and click Start. It reads a scanned letter (PNG) and a 4-page PDF that mixes typed and scanned pages:

{
  "startUrls": [
    {"url": "https://api.apify.com/v2/key-value-stores/nEPO9kOfmRIEPfuSU/records/scan-letter.png?signature=1bcKYhjnnj39zb3oJzzcL"},
    {"url": "https://api.apify.com/v2/key-value-stores/nEPO9kOfmRIEPfuSU/records/mixed.pdf?signature=1FMu7z7IhbDt9yOUxaxS"}
  ],
  "languages": ["eng"]
}

You get 5 rows (one per page) in about 15 seconds: 3 pages read with OCR and charged, and 2 typed PDF pages read from the text layer for free.

What data can I get?

Group	Fields
Source	source, fileName, fileType (pdf, png, jpeg, webp, tiff, gif, bmp, heic, avif), page, pageCount, languages
Text	text (always), markdown (Markdown output), lines (Boxes output: each line and word with a pixel box [x0, y0, x1, y1] and a confidence)
Result	method (ocr, text-layer, blank or error), charged, confidence (0-100), wordCount, characters
Clean-up	rotation (degrees turned upright), skewDegrees, enhanced, scale
Image	width, height (pixels), dpi (PDF render resolution)
Files and errors	searchablePdfUrl (with Also make a searchable PDF), error

One real row (run on 2026-10-18, page 3 of the sample PDF, text and source shortened):

{
  "source": "https://api.apify.com/v2/key-value-stores/nEPO9kOfmRIEPfuSU/records/mixed.pdf?signature=…",
  "fileName": "mixed.pdf",
  "fileType": "pdf",
  "page": 3,
  "pageCount": 4,
  "method": "ocr",
  "charged": true,
  "text": "Signed minutes\n\nNext steps: order the timber by the end of the month, ask the council about a second water tap, …",
  "confidence": 96.5,
  "wordCount": 64,
  "characters": 267,
  "languages": ["eng"],
  "rotation": 0,
  "skewDegrees": 0,
  "enhanced": false,
  "scale": 1,
  "width": 2481,
  "height": 3508,
  "dpi": 300,
  "searchablePdfUrl": null,
  "error": null
}

The dataset has three views (Pages, Markdown, Page details) and exports to JSON, CSV, Excel or HTML. The run's SUMMARY record lists every file with its OCR, text-layer, blank, failed and skipped page counts.

What it handles:

• Files: PNG, JPG, WebP, TIFF (every page of a multi-page TIFF), GIF, BMP, HEIC/HEIF (iPhone photos) and AVIF, plus scanned, mixed or born-digital PDFs, from links, an upload, or an Apify key-value store. Google Drive and Dropbox share links are turned into direct downloads.
• Only the pages that need it: in a mixed PDF, typed pages are read from their text layer (exact and free) and only scanned pages are OCR'd. Set Which PDF pages to OCR to every page to re-read them all.
• Clean-up: sideways and upside-down pages are turned upright, tilted scans (up to 15 degrees) are levelled, shadows on phone photos and faded print are evened out, small print is enlarged, and hard pages get a second pass. Each row says what was done.
• Layout: Markdown output keeps headings, paragraphs and lists, and two-column pages come out column by column. A narrow column such as a receipt is read as one block; set Page layout to force it either way.
• 20 languages, up to 4 per run: English, Spanish, French, German, Portuguese, Italian, Dutch, Polish, Russian, Ukrainian, Turkish, Arabic, Hindi, Chinese (Simplified and Traditional), Japanese, Korean, Vietnamese, Indonesian and Thai. Codes (deu), short codes (de) and names (German) all work.

How accurate is it? The engine is Tesseract 5 with our own clean-up in front of it. On 28 test pages we made ourselves (clean, tilted and noisy scans, a fax, phone photos with shadows, faded and tiny print, six languages), the average character error was 0.0% on the 20 standard pages and 4.0% on the 8 hard ones, against 5.3% and 39% for plain Tesseract without the clean-up. We tuned the clean-up on these pages, so treat the numbers as a guide, not a promise.

How much does it cost?

You pay per page read with OCR that returned text. There is no charge for Apify platform usage on top.

Row	Free and Starter	Scale	Business
OCR page	$3.00 per 1,000	$2.25 per 1,000	$1.65 per 1,000

Plus $0.00005 per run start (charged per GB of memory, so $0.0002 at the default 4 GB). Apify's free $5 monthly credit covers about 1,600 OCR pages. Free-plan runs OCR up to 100 pages each (text-layer pages don't count); any paid Apify plan removes the cap.

What's free: PDF pages read from their text layer, blank pages and pictures with no text, pages and files that fail (bad links, web pages instead of files, unsupported formats, password-protected PDFs), duplicate links, and pages not reached because of a cap or the run's time limit. To cap a run's spend, set Maximum cost per run in the run options, or Max OCR pages per run: the run stops before going over and SUMMARY says how many pages were left.

Limits, stated up front

• Handwriting is not read well. Typed and printed text only.
• Tables come out as lines of text, not as table cells.
• Low-resolution faxes (around 100 dpi with broken letters) still come back with mistakes; scan at 200 dpi or more when you can.
• Strong perspective (a page photographed at a steep angle) is not flattened. Photograph from above.
• Password-protected PDFs are skipped with a message. Remove the password first.
• Up to 1,000 files per run. Files up to 100 MB by default (adjustable to 500 MB). Max pages per file reads only the first N pages of each PDF or TIFF.
• Speed: 2 pages at a time at the default 4 GB memory, one more for each extra 2 GB, up to 8. The quick start's 5 pages took 7 seconds of run time. If the run's time limit is reached, pages already read are delivered and the rest are listed in SUMMARY, not charged.
• Word boxes are in pixels of the original image; for PDF pages, pixels at the render resolution in dpi (300 by default), so x / dpi * 72 gives PDF points.

Schedules

Save a task that points at a key-value store or a list of links, then add a schedule under Schedules in Apify Console to OCR new scans as they arrive. Each run reads every file you give it, so point it at the new files only.

Integrations: API, n8n, Make and AI agents

API. Run it and get the rows back in one call:

curl -X POST "https://api.apify.com/v2/acts/locaihost~ocr/run-sync-get-dataset-items?token=YOUR_APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"startUrls": [{"url": "https://example.com/scan.png"}], "languages": ["eng", "deu"]}'

Apify's API clients for Python and JavaScript work the same way.

n8n, Make, Zapier, Google Sheets, Slack, webhooks. Apify's built-in integrations send each run's dataset wherever you need it.

AI agents (MCP). Add this Actor to Claude, Cursor or any MCP client with https://mcp.apify.com/?tools=locaihost/ocr, then ask something like "Read this scanned invoice and tell me the total and the due date."

FAQ

How much does it cost to OCR a scanned PDF?

$3.00 per 1,000 OCR pages on the Free and Starter plans, less on Scale and Business. Typed pages, blank pages and failures are free. Apify's free $5 monthly credit covers about 1,600 OCR pages.

Can I use it through the API or from Python?

Yes. Use the run-sync-get-dataset-items call above, or Apify's Python and JavaScript clients. Every run's rows are also available as JSON, CSV or Excel from its dataset.

Can I use it from an AI agent?

Yes. Through the Apify MCP server (https://mcp.apify.com/?tools=locaihost/ocr) any MCP client can send an image or PDF link and read the text back.

Is it legal to OCR documents this way?

The Actor reads only the files you supply. It doesn't crawl or collect anything else, everything runs inside your own run, and the files and results stay in your run's storage. Documents can contain personal data: you are responsible for having the right to process them and for handling that data lawfully.

Why are some fields empty?

markdown and lines appear only with Markdown or Boxes output. Text-layer pages have no confidence, rotation or skewDegrees, because they were not OCR'd. A blank page has empty text and method: "blank". A failed file has one row with method: "error" and the reason in error.

Does it work on Word or Excel files?

No: send PNG, JPG and other images, or PDFs. Word, PowerPoint and Excel files are reported as unsupported and not charged; the Document to Markdown Actor below converts them.

More from locaihost data

• Document to Markdown converter: PDFs with text, Word, PowerPoint and Excel files to clean Markdown.
• Website screenshot API: capture web pages as images or PDFs.
• YouTube transcript scraper: video captions as text for the same pipeline.

Did it help?

If this saved you time, a quick rating on the Store helps other people find it.

Support

Something not working, or a language or format you need? Open an issue on the Issues tab: we answer issues within 24 hours.