Image OCR Scraper

Extract text from public image URLs with local OCR. This actor is designed for developers, researchers, and AI agents that need a simple image URL → structured OCR result tool for screenshots, scanned documents, menus, labels, receipts, and other images they are authorized to process.

What it does

• Downloads each supplied http or https image URL with retries.
• Runs Tesseract OCR locally in the actor using the selected language.
• Returns one dataset item per image with recognized text, confidence, content type, byte count, timestamp, and status.
• Keeps processing after an individual download or OCR failure, so mixed batches produce useful partial results.
• Charges PPE once per successfully processed image (ocr-image). Failed downloads and OCR errors are not charged.

What it does not do

This actor does not log in, bypass paywalls or bot protection, crawl image galleries, discover image URLs from websites, or process private images without a reachable URL. Use only images that you have permission to access and process. Image URLs should point directly to image content; HTML pages containing an image are not supported.

Input

{
  "imageUrls": [
    "https://example.com/receipt.png",
    "https://example.com/menu.jpg"
  ],
  "language": "eng",
  "requestTimeoutSeconds": 30
}

imageUrls accepts up to 25 URLs per run. Supported language presets are eng, deu, fra, and spa. The first run downloads the selected Tesseract language data, so startup can be longer than later images in the same run. Images are limited to 25 MB each.

Output

Each image produces a row like this:

{
  "status": "succeeded",
  "imageUrl": "https://example.com/receipt.png",
  "language": "eng",
  "text": "TOTAL $42.00",
  "confidence": 93.41,
  "bytesDownloaded": 184220,
  "contentType": "image/png",
  "processedAt": "2026-08-13T00:00:00.000Z"
}

Failed rows have status: "failed" and an error message. A not_charged row means the actor's pay-per-event limit was reached before returning an OCR result.

MCP and automation use

Use this actor when an agent already has direct image URLs and needs text extraction. It is intentionally narrow: pass URLs, choose a language, and read structured rows from the dataset. For a website-wide image crawl, pair it with a URL-discovery actor and pass the discovered direct image URLs into this actor.

Pricing

Pay per usage: one successful OCR image is one billable event. The default event price is $0.002 per image; your final Store price and account settings control the amount charged in production.

Limitations

OCR accuracy depends on image resolution, contrast, rotation, handwriting, typeface, and language. Confidence is Tesseract's estimate, not a guarantee. Pre-crop or enhance difficult images before processing when accuracy is important.