OCR built for receipts. Photos, scans, thermal paper, faded ink — the API reads them all and returns structured JSON. Merchant, items, tax, total, payment method. Skip the regex pipeline.
Send a receipt image and a schema. The API runs OCR, understands the receipt layout, and returns the fields you asked for — typed and ready to insert into your database.
curl -X POST \
https://api-parse.conversiontools.io/v1/extract \
-H "Authorization: Bearer YOUR_API_KEY" \
-F "file=@receipt.jpg" \
-F 'schema={
"merchant": "string",
"merchant_address": "string",
"transaction_date": "date",
"items": [{
"name": "string",
"price": "number"
}],
"subtotal": "number",
"tax": "number",
"total": "number",
"payment_method": "string"
}'{
"status": "completed",
"pages": 1,
"data": {
"merchant": "Trader Joe's",
"merchant_address": "1800 N Vermont Ave, LA",
"transaction_date": "2026-02-20",
"items": [
{ "name": "Org Bananas", "price": 0.99 },
{ "name": "Sourdough Bread", "price": 4.49 }
],
"subtotal": 25.32,
"tax": 2.41,
"total": 27.73,
"payment_method": "Visa ****4829"
}
}Tilted shots, hand-held photos, indoor lighting — the model is trained on real-world receipt captures, not flat-bed scanner ideals. Good enough beats picture-perfect every time.
Light gray ink, narrow columns, monospaced print — the OCR pipeline boosts contrast on thermal paper before reading. Faded receipts still degrade gracefully.
Most receipt OCR APIs return a flat string of line items. Parse returns a typed array, so each item has its own name and price field — ready to load into a transactions table.
The OCR engine handles common real-world conditions: low light, slight blur, perspective distortion, glare, and faded thermal paper. The model uses context (it knows the document is a receipt) to disambiguate characters that pure character-level OCR would miss.
Yes. Define line_items as an array of objects in your schema with fields like description, quantity, unit_price, amount. The API extracts each item as a structured row — even when the receipt formatting is irregular or items wrap across lines.
JPEG, PNG, WebP, TIFF, and PDF. Most expense apps capture as JPEG; the API processes any of these without conversion.
Yes. The model handles multilingual receipts including Latin scripts, CJK (Chinese/Japanese/Korean), Cyrillic, Arabic, and others. You can also localize date and number formats by declaring them in your schema.
Fresh thermal receipts read very accurately. Faded thermal paper (older than ~6 months) drops in accuracy because the print itself is degrading — there's nothing OCR can recover that the human eye can't. Tip: scan or photograph receipts within a month of purchase for best results.
Free tier covers your first 100 receipts a month. No credit card to start.