Image OCR Text Extractor: extract text from images

🔤 What is Image OCR?

Extract text from images in bulk: scans, screenshots, letters and phone photos become searchable copy. Give it links or uploaded files and receive one row per picture with clean Markdown (headings, lists and table rows kept) and the raw characters. Optical character recognition runs inside the run with an open-source library (Apache-2.0), several scripts at once for mixed documents, and nothing is sent to an outside OCR service or AI model.

• Document digitisation: turn scanned letters, forms and book pages into searchable text
• Receipts and bills: read statements and till slips before you parse amounts and dates
• RAG and AI agents: feed the Markdown of screenshots and scans to your knowledge base or LLM
• Screenshots: pull the text out of app, chat or social media captures in bulk
• Layout and redaction: use the word boxes to find, highlight or blank out words on the image

✨ Why use this Actor?

• ✅ 29 languages: English, German, French, Spanish, Italian, Portuguese, Dutch, Polish, Russian, Arabic, Hindi, Japanese, Korean, Chinese and more, several per image
• ✅ Clean Markdown: large short lines become headings, bullets and numbered items turn into lists, wrapped and hyphenated breaks are joined, tables and addresses keep their layout
• ✅ Pixel boxes: x, y, width, height and confidence for every word, with its block, paragraph and line
• ✅ Noise filter: minConfidence drops words Tesseract is unsure of; 50 to 60 cleans most phone photos
• ✅ Private: OCR runs inside the run; nothing is sent to an outside service
• ✅ Fair price: 0.002 USD per image; images that fail are not charged

📦 What data can you get with Image OCR?

📝 Plain text	Ⓜ️ Markdown	🔲 Word boxes in pixels
🎯 Confidence per word and image	🔢 Word and line count	🌍 Languages used
🖼️ Image format	🔗 Source URL or key

🚀 How to use Image OCR

Image OCR works in a few clicks, no coding needed:

1. Create a free Apify account (the free plan includes 5 USD of monthly credit).
2. Open Image OCR and click Try for free.
3. Paste the image links (or keys of uploaded images) and pick the languages in the images.
4. Click Start and wait for the run to finish (most runs take seconds).
5. Download your data as JSON, CSV, Excel, XML or HTML, or send it straight to Google Sheets, Make, Zapier or your AI agent.

⬇️ Input

Give imageUrls (PNG, JPEG, WebP, BMP, PBM/PGM/PPM) or storeKeys; the defaults read English and return text, Markdown and word boxes.

Languages

Pick every language that appears in the images, for example eng, deu and fra for a European page; each extra one slows the run a little.

Smaller rows

Switch off includeWords and includeText when you only need the Markdown.

Limits

Up to 500 images per run; files over maxSizeMb (20 MB by default) are refused before download, anything above 100 million pixels is skipped, the run stops once 3 pictures hit the 5-minute time limit, and downloads are retried maxRetries times on network errors, 429 and 5xx.

Field	Type	What it does	Default
Image URLs (imageUrls)	array	Direct links to images (PNG, JPEG, WebP, BMP, PBM/PGM/PPM).	["https://tesseract.projectnaptha.com/im
Uploaded files (storeKeys)	array	Keys of images stored in this run's default key-value store (upload them with the API before the run starts).
Languages (languages)	array	Languages in the images.	["eng"]
Include words with boxes (includeWords)	boolean	Every word with its pixel box (x, y, width, height), confidence and its block, paragraph and line.	true
Include plain text (includeText)	boolean	Plain text with the line breaks of the image.	true
Min word confidence (minConfidence)	integer	Words Tesseract is less sure of (0 to 100) are dropped from text, Markdown and words.	0
Max file size (MB) (maxSizeMb)	integer	Larger files are refused before they are downloaded.	20
Retries (maxRetries)	integer	Retries per download on network errors, 429 and 5xx, with exponential backoff.	3

Input example

{
    "imageUrls": [
        "https://tesseract.projectnaptha.com/img/eng_bw.png",
        "https://raw.githubusercontent.com/naptha/tesseract.js/master/benchmarks/data/testocr.png"
    ],
    "languages": [
        "eng"
    ]
}

⬆️ Output

Every image becomes one row with its text, Markdown, line and token totals and mean confidence, plus the word boxes when includeWords is on. The run summary (OUTPUT) counts images processed, words, images without text and per-image errors; the run fails only when no image could be read.

Output example

One row of the dataset (a bank statement, Markdown only, a real run of 2026-10-15, shortened):

{
    "source": "https://raw.githubusercontent.com/naptha/tesseract.js/master/tests/assets/images/bill.png",
    "format": "png",
    "languages": "eng",
    "markdown": "FIRST CHEQUING\n\nLine of Credit 100,000.00 Rate 4.2000\n\nDate Description Number Debits Credits Balance\n31Jul2018 Forward 99,878.08 -\n01Aug2018 Clearing Cheque 4987 36.07 99,914.15 -\n…",
    "wordCount": 93,
    "lineCount": 15,
    "confidence": 86
}

View the results as a table or JSON in the Output tab and download them as JSON, CSV, Excel, XML, HTML or RSS. Pick the fields you need first, or take the whole dataset.

💰 How much does Image OCR cost?

Image OCR is priced per event: 2 USD per 1,000 images (0.002 USD per image), with no monthly rental. Images that fail (download, format, size) are not charged.

You get	You pay
100 images read	0.2 USD
1,000 images read	2 USD
10,000 images read	20 USD
100,000 images read	200 USD

💡 Apify's free plan gives you 5 USD of credit every month, enough for about 2,500 images read at no cost. Set a spending limit per run and the Actor stops there.

⚖️ Image OCR vs. the alternatives

	Image OCR	Cloud OCR APIs	Copy by hand
Text, Markdown and word boxes	✅	text and boxes	❌
Images stay private	✅	❌	✅
No API key or cloud account	✅	❌	✅
Price per 1,000 images	2 USD	1.5 USD and up, plus setup	hours of work
Handwriting	❌	often	✅

🔌 Integrations, API and MCP

Connect Image OCR to your AI agent (MCP)

Give Claude, ChatGPT, Cursor or any MCP client direct access to Image OCR through the Apify MCP server. Add this URL to its settings:

https://mcp.apify.com/?actors=bepolabs/image-ocr

For Claude Desktop, add it under mcpServers in claude_desktop_config.json:

{
    "mcpServers": {
        "image-ocr": {
            "command": "npx",
            "args": [
                "-y",
                "@apify/actors-mcp-server",
                "--actors",
                "bepolabs/image-ocr"
            ],
            "env": {
                "APIFY_TOKEN": "<your Apify API token>"
            }
        }
    }
}

Then just ask: "Read the text in these three receipt photos and give it to me as Markdown."

Run it from your code

Python (pip install apify-client):

from apify_client import ApifyClient

client = ApifyClient("<YOUR_APIFY_TOKEN>")
run = client.actor("bepolabs/image-ocr").call(run_input={"imageUrls":["https://tesseract.projectnaptha.com/img/eng_bw.png","https://raw.githubusercontent.com/naptha/tesseract.js/master/benchmarks/data/testocr.png"],"languages":["eng"]})
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item)

JavaScript (npm install apify-client):

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: '<YOUR_APIFY_TOKEN>' });
const run = await client.actor('bepolabs/image-ocr').call({"imageUrls":["https://tesseract.projectnaptha.com/img/eng_bw.png","https://raw.githubusercontent.com/naptha/tesseract.js/master/benchmarks/data/testocr.png"],"languages":["eng"]});
const { items } = await client.dataset(run.defaultDatasetId).listItems();
console.log(items);

No-code integrations

Send results to Google Sheets, Slack, Airtable, HubSpot, Make, Zapier, n8n and more from the Integrations tab, start runs on a schedule, or trigger your own endpoint with a webhook when a run finishes. The Apify API gives you full control over runs and datasets.

❓ FAQ

Which images read well?

Scans and screenshots at 150 to 300 dpi read at 90 or more mean confidence. Phone photos work when the page is flat and in focus; set minConfidence to 50 to 60 to drop noise.

Can it read handwriting?

Not reliably: Tesseract is built for printed text.

Does it read PDFs?

No, images only. For files with a text layer use PDF to Text and Markdown; scanned documents need their pages exported as images first.

Are my images stored or sent elsewhere?

No. Each image is read inside the run with Tesseract; nothing is sent to an outside OCR service or AI model.

What if one image fails?

That image gets an error line in the OUTPUT summary and is not charged; the other images go on. GIF and TIFF are refused with a message to convert them to PNG or JPEG.

Is it legal to use Image OCR?

The Actor reads only the images you give it, inside the run. Make sure you may process them; copyright, confidentiality and any personal data in the images stay your responsibility.

What if something is missing or does not work?

Open an issue on the Issues tab; we read each one and reply. Feature requests are welcome too.

💬 Your feedback

Tell us what Image OCR should add on the Issues tab or leave a review; your notes decide what we work on next.

🔗 Related Actors

Actor	What it does
📄 PDF to Text and Markdown	PDFs to clean text and Markdown, per page, per document or as RAG chunks, with columns in reading order and headers and footers removed.
🎙️ Audio and Video Transcriber	Whisper transcripts of audio and video files: full text, timed segments, SRT and WebVTT. No API key.
📝 Website to Markdown	Any web page or docs site to clean Markdown for LLMs, RAG and AI agents, with headings, links and ready-made chunks.
📝 YouTube Transcripts	Full text and timed segments of the captions of any YouTube video, ready for summaries, search and LLMs.
🎓 OpenAlex Research Papers	Research papers in bulk from OpenAlex: search, DOI or citing works to clean rows with citations, abstracts and free PDF links.
🎙️ Podcast Episodes	Search terms, Apple Podcasts ids or RSS feeds to every show and its episodes, with audio links, dates and durations.
📰 RSS Feed Reader	Any RSS, Atom or JSON feed, or just a website URL, to one clean list of articles, with full text on request.
🏛️ SEC EDGAR Filings	Tickers or CIK numbers to SEC filings: 10-K, 10-Q, 8-K, Form 4 and any other form, with document links.