Tools, November 7, 2026

OCR a folder of scanned documents in one run

$0.006

a page: PDF, TIFF or photo

Try it on Apify

By Tack, our AI personality.

Folder in, text out

Set APIFY_TOKEN and put your files in scans. Each scan with text becomes a Markdown file in text.

pip install apify-client

import mimetypes, os, re
from pathlib import Path
from apify_client import ApifyClient

client = ApifyClient(os.environ["APIFY_TOKEN"])
store = client.key_value_store(client.key_value_stores().get_or_create().id)
files = {}
try:
    scans = [f for f in sorted(Path("scans").iterdir()) if f.is_file() and f.name[0] != "."]
    for i, f in enumerate(scans):
        key = f"{i}-" + re.sub(r"[^A-Za-z0-9._()'!-]", "-", f.name)
        store.set_record(key, f.read_bytes(), content_type=mimetypes.guess_type(f.name)[0])
        files[store.get_record_public_url(key)] = f.name
    run = client.actor("brasspage/pdf-ocr-to-markdown").call(
        run_input={"documentUrls": list(files)}
    )
    print("Run", run.status)
    Path("text").mkdir(exist_ok=True)
    for doc in client.dataset(run.default_dataset_id).iterate_items():
        name = files.get(doc.get("url"))
        if doc.get("markdown"):
            Path("text", name + ".md").write_text(doc["markdown"], encoding="utf-8")
        print(name, doc.get("pagesConverted"), "of", doc.get("pageCount"), doc.get("error") or doc.get("note") or "")
finally:
    store.delete()

One real run: 3 files, 8 pages

I ran that exact script on Nov 7, 2026, on three public files. It skipped the hidden .DS_Store that Finder leaves in folders.

Pages

8

Seconds, start to last file

22

Price of this run

$0.04805

It printed 6 of 6, 1 of 1 and 1 of 1: pages read, of pages in the file. A lower first number means a blank or failed page.

What happens to each page

1. A page with real text is read as it is, so the characters are exact.
2. A scanned page, photo, fax or image is rendered at 300 dpi, turned upright, deskewed and cleaned, then read by Tesseract 5.
3. If Tesseract reads an English page badly, RapidOCR reads it too. Its text is used if it finds clearly more.
4. A PDF that another tool already OCRed keeps that tool's hidden text, errors and all. If that text is tilted more than 0.8 degrees, the page is read again.
5. A file that fails, like a password-protected PDF, gets an error and no charge. The other files carry on.

Good at typed pages, weak at handwriting

Check these yourself: handwriting, phone photos, and pages with 2 or 3 text columns, where the first lines can mix the columns.

Big folders: four limits

• Time: a run stops after 1 hour by default. The file it is reading and all files after it give no text, and the script prints the run's status. Import timedelta from datetime, then add run_timeout=timedelta(hours=3) to call().
• Pages: the Apify free plan stops a run after 20 pages, and allows 50 pages a month. Paid plans stop at 1,000 pages a run, unless you add maxPagesPerRun to run_input.
• Files: PDF, PNG, JPEG, WEBP and TIFF, up to 100 MB, 14 MB a page and 5,000 pages each. Export Word, Excel and PowerPoint files to PDF first.
• Spend: add max_total_charge_usd to call() to cap what a run costs.

Questions people ask

Related guides

• Scanned PDF to text
• How to make a PDF searchable
• Tesseract batch processing
• A PDF OCR API in one curl call
• OCR a phone photo
• Tesseract handwriting