Can I get the document's raw structure instead of extracted fields?

Yes. This OCR API returns validated fields (invoice_number, total, dates…). If you'd rather have the whole document as structured Markdown/JSON — layout, tables and reading order — to build your own extraction or an LLM/RAG pipeline, use the Document Parsing API . Same auth, same upload; different output.

Can I use the OCR API from Python?

Yes — plain requests, no SDK required: one requests.post() to upload the document and one requests.get() to read the parsed JSON (see the Python tab in the quickstart). Any language that speaks HTTPS works the same way.

What file types does the OCR API accept?

PDFs (including long multi-page files), JPEG, PNG and TIFF images, and ZIP batches. Mixed or bundled files are split and classified automatically before extraction.

Does the OCR API handle receipt scanning?

Yes — receipt scanning is a pre-trained type: merchant, date, tax breakdown, line items and payment method come back as typed fields from photos, scans or PDFs. Same endpoint and flow as any other document, and crumpled thermal-paper receipts are exactly what the model was trained on.

How accurate is the extraction?

Every field comes back typed and validated before delivery. Accuracy is measured against ground truth on your own documents — see the evaluation methodology — and it is SLA-backed: you only pay for fields extracted correctly.

What's the typical latency? Are there rate limits?

Around ten seconds from upload to structured JSON for a typical document. Processing is asynchronous — webhooks instead of polling — and batches run through queues with autoscaling, so volume spikes don't need client-side throttling.

What does it cost per document?

Usage-based, per page — no seats, no setup fees. Your first 500 pages are free, no credit card. See the pricing page for volume tiers.

How long is document data retained?

As long as you need — or not at all: zero-retention mode purges documents and extracted data on delivery. SOC 2 Type II, ISO 27001 and GDPR compliant; see the zero-retention policy .

Document OCR API, built for developers

OCR any PDF document or image through one endpoint and get structured, validated JSON back — invoices, utility bills, mortgage documents, IDs or any schema you define. Auth is one header; results arrive in seconds via webhook.

Start for free Read the docs

500 pages free
No credit card

# Upload a document — any PDF or image
curl -X POST https://api.invofox.com/v1/ingest/uploads \
  -H "x-api-key: $INVOFOX_API_KEY" \
  -F "[email protected]" \
  -F 'info={"type":"invoice"}'

# → the document.processed webhook delivers:
{
  "type": "invoice",
  "data": {
    "invoiceNumber": "INV-2026-0841",
    "provider": {
      "name": "Acme Supplies GmbH",
      "taxId": "DE811570831"
    },
    "totalAmount": 1804.28
  }
}

import requests

# Upload a document — any PDF or image
requests.post(
    "https://api.invofox.com/v1/ingest/uploads",
    headers={"x-api-key": "YOUR_API_KEY"},
    files={"files": open("invoice.pdf", "rb")},
    data={"info": '{"type": "invoice"}'},
)

# → the document.processed webhook delivers:
{
  "type": "invoice",
  "data": {
    "invoiceNumber": "INV-2026-0841",
    "provider": {
      "name": "Acme Supplies GmbH",
      "taxId": "DE811570831"
    },
    "totalAmount": 1804.28
  }
}

import { openAsBlob } from "node:fs"; // Node 20+

// Upload a document — any PDF or image
const form = new FormData();
form.append("files", await openAsBlob("invoice.pdf"), "invoice.pdf");
form.append("info", JSON.stringify({ type: "invoice" }));
await fetch("https://api.invofox.com/v1/ingest/uploads", {
  method: "POST",
  headers: { "x-api-key": process.env.INVOFOX_API_KEY },
  body: form,
});

// → the document.processed webhook delivers:
{
  "type": "invoice",
  "data": {
    "invoiceNumber": "INV-2026-0841",
    "provider": {
      "name": "Acme Supplies GmbH",
      "taxId": "DE811570831"
    },
    "totalAmount": 1804.28
  }
}

OCR a PDF document to structured JSON

This is the invoice from the quickstart above, in full — line items, dates, IBAN and totals that actually add up, ready for your ERP.

Fields shown are the default invoice schema — define your own fields for any document type.

GET /documents/{id} json

 {
  "_id": "68a1f30b2c9e4d0012f7a1b3",
  "type": "invoice",
  "data": {
    "invoiceNumber": "INV-2026-0841",
    "issueDate": "2026-07-02",
    "dueDate": "2026-08-01",
    "provider": {
      "name": "Acme Supplies GmbH",
      "taxId": "DE811570831",
      "iban": "DE89370400440532013000"
    },
    "lines": [
      {
        "description": "Thermal paper rolls 80x80",
        "quantity": 40,
        "unitPrice": 21.90,
        "amount": 876.00
      },
      {
        "description": "Label printer ZD421",
        "quantity": 2,
        "unitPrice": 307.85,
        "amount": 615.70
      },
      {
        "description": "Shipping & handling",
        "quantity": 1,
        "unitPrice": 24.50,
        "amount": 24.50
      }
    ],
    "subtotal": 1516.20,
    "taxRate": 0.19,
    "taxAmount": 288.08,
    "totalAmount": 1804.28,
    "currency": "EUR"
  }
}

Limits, latency and pricing

The numbers you'd otherwise dig out of the docs.

~10s typical latency

From upload to structured JSON

Processing is asynchronous — subscribe to the document.processed webhook instead of polling.
1→N classifier + splitter

Batches and mixed files welcome

Send a zipped batch or one messy multi-document PDF — the splitter separates it and the classifier routes each document to the right extractor.
Any document type

Not just invoices

Utility bills, mortgage packets, IDs, delivery notes — pre-trained types or your own schema. Layout-independent by design, no templates.
500 pages free

Usage-based pricing

Per-page pricing that scales linearly — no seats, no setup fees. Start free, no credit card.

Every document type, one API

Pre-trained types extract data from any PDF document or image out of the box — or define your own schema for anything else. Each linked card has its own deep-dive page.

Invoices

Extract invoice number, dates, totals, vendor details, line items and more.
Purchase orders

Extract buyer & supplier details, order date, delivery date, items ordered and more.
Utility bills

Extract account holder, account number, billing period, usage details and more.
Receipts

OCR receipt scanning at scale — merchant, date, VAT, line items and payment method from photos or PDFs.
Payslips

Extract deductions, payment details, tax IDs and more.
Closing disclosures

Extract loan terms, closing costs, cash to close, interest rate, escrow details and more.
Bank statements

Extract account holder, balances, transactions, dates, IBAN and more.
Tax forms

Extract taxpayer details, taxable base, withholdings, tax due, reference numbers and more.
Expense reports

Extract merchant, category, amount, date, tax and reimbursable totals and more.

Also pre-trained: Bill of lading · Checks · Pro-forma invoices · Lohnkonto — or define your own schema for anything else →

Why a specialized OCR API beats DIY

What you'd have to build around Tesseract or a hyperscaler API is already inside.

Validated JSON, not raw text

OCR engines return characters; hyperscalers return boxes. This API returns typed, validated fields with your business rules already applied.
Any layout, zero templates

Trained on real-world document diversity — skewed scans, photos, 100-page PDFs. No per-vendor templates to build or maintain.
Document ops included

Retries, async queues, webhooks, classification and splitting ship with the API — the infrastructure DIY pipelines spend months rebuilding.

Benchmarked, not claimed: GPT-4o vs Claude vs Invofox for document parsing — with code samples →

OCR API FAQ

~/invofox / faq.json

// questions 9

1 {

2 ··"question": "How do I OCR a PDF to structured JSON?",

3

4 ··"answer": "Upload the PDF to the ingest endpoint with its document type, subscribe to the document.processed webhook, and read the extracted fields from the document endpoint. The whole flow is shown in the quickstart above in cURL, Python and Node — no OCR engine to run, no post-processing to write."

5 }

How-to howto.json
1 {

2 ··"question": "Can I get the document's raw structure instead of extracted fields?",

3

4 ··"answer": "Yes. This OCR API returns validated fields (invoice_number, total, dates…). If you'd rather have the whole document as structured Markdown/JSON — layout, tables and reading order — to build your own extraction or an LLM/RAG pipeline, use the Document Parsing API. Same auth, same upload; different output."

5 }

Parsing parsing.json
1 {

2 ··"question": "Can I use the OCR API from Python?",

3

4 ··"answer": "Yes — plain requests, no SDK required: one requests.post() to upload the document and one requests.get() to read the parsed JSON (see the Python tab in the quickstart). Any language that speaks HTTPS works the same way."

5 }

Python python.json
1 {

2 ··"question": "What file types does the OCR API accept?",

3

4 ··"answer": "PDFs (including long multi-page files), JPEG, PNG and TIFF images, and ZIP batches. Mixed or bundled files are split and classified automatically before extraction."

5 }

Formats formats.json
1 {

2 ··"question": "Does the OCR API handle receipt scanning?",

3

4 ··"answer": "Yes — receipt scanning is a pre-trained type: merchant, date, tax breakdown, line items and payment method come back as typed fields from photos, scans or PDFs. Same endpoint and flow as any other document, and crumpled thermal-paper receipts are exactly what the model was trained on."

5 }

Receipts receipts.json
1 {

2 ··"question": "How accurate is the extraction?",

3

4 ··"answer": "Every field comes back typed and validated before delivery. Accuracy is measured against ground truth on your own documents — see the evaluation methodology — and it is SLA-backed: you only pay for fields extracted correctly."

5 }

Accuracy accuracy.json
1 {

2 ··"question": "What's the typical latency? Are there rate limits?",

3

4 ··"answer": "Around ten seconds from upload to structured JSON for a typical document. Processing is asynchronous — webhooks instead of polling — and batches run through queues with autoscaling, so volume spikes don't need client-side throttling."

5 }

Latency latency.json
1 {

2 ··"question": "What does it cost per document?",

3

4 ··"answer": "Usage-based, per page — no seats, no setup fees. Your first 500 pages are free, no credit card. See the pricing page for volume tiers."

5 }

Pricing pricing.json
1 {

2 ··"question": "How long is document data retained?",

3

4 ··"answer": "As long as you need — or not at all: zero-retention mode purges documents and extracted data on delivery. SOC 2 Type II, ISO 27001 and GDPR compliant; see the zero-retention policy."

5 }

Security retention.json

howto.json

1 {

2 ··"question": "How do I OCR a PDF to structured JSON?",

4 ··"answer": "Upload the PDF to the ingest endpoint with its document type, subscribe to the document.processed webhook, and read the extracted fields from the document endpoint. The whole flow is shown in the quickstart above in cURL, Python and Node — no OCR engine to run, no post-processing to write."

5 }

How-to howto.json

Something still unclear? Read the docs

Ship document extraction this afternoon

One endpoint, one header, validated JSON back — a typical integration takes under a day. Building with an AI agent? Point it at the integration guide and let it write the code.

Start for free Integrate with an AI agent

500 pages free · No credit card · Works from Claude Code, Cursor or any MCP client

Document OCR API, built for developers

OCR a PDF document to structured JSON

Limits, latency and pricing

From upload to structured JSON

Batches and mixed files welcome

Not just invoices

Usage-based pricing

Every document type, one API

Invoices

Purchase orders

Utility bills

Receipts

Payslips

Closing disclosures

Bank statements

Tax forms

Expense reports

Why a specialized OCR API beats DIY

Validated JSON, not raw text

Any layout, zero templates

Document ops included

OCR API FAQ

Ship document extraction this afternoon