Built for Claude Code and OpenAI Codex workflows

Build apps that can read the documents your customers send.

Bank statements, photographed receipts, faxed forms, handwritten notes. OCRWell turns any of them into plain text or typed JSON your code can use, in one API call.

Try it on a document Try the live demo See how it works 200 OCR + 50 extraction pages / month. Free forever, no card.

Input · Photographed receipt → structured items 2 / 3

Tap image to zoom. Use arrows for more examples.

receipt.json

json

{
  "data": {
    "merchant": "GREENGATE GROCERS",
    "date": "11/04/2026",
    "items": [
      {
        "name": "Flat White",
        "price": 3.8
      },
      {
        "name": "Sourdough Loaf",
        "price": 4.5
      },
      {
        "name": "Oat Milk 1L",
        "price": 2.2
      },
      {
        "name": "Bananas 1kg",
        "price": 1.9
      }
    ],
    "subtotal": 12.4,
    "vat": 2.48,
    "total": 14.88
  }
}

Run OCR on your own document

Works on PDFs, phone photos, scans, and handwriting. No templates, no fine-tuning, no training.
Plain text, or data shaped to your own JSON schema, ready for the next step in your pipeline.
Ships as a skill for Claude Code, Codex, Gemini CLI, Antigravity, and Cursor. Your coding agent writes the integration.

How it works

Your agent writes the integration. You review the diff.

Install the OCRWell skill once, set an API key, and ask your agent to build the feature. The skill ships the rules it would otherwise guess at, so the code that lands in your project uses the API correctly the first time.

The call itself is one function.

extract.py

python

from ocrwell import client

result = client.extract(
    file=open("receipt.png", "rb"),
    schema={"total": "number", "items": "array"},
)

Install in Claude Code

Uses the built-in plugin marketplace, so you get versioning and /plugin update for free.

claude-code

text

/plugin marketplace add ocrwell/skill
/plugin install ocrwell@ocrwell

Install in Codex CLI (or Claude Code)

One npx target per agent. Writes the SKILL.md to the location your agent reads from.

terminal

bash

$ npx @ocrwell/skill install --codex
$ export OCRWELL_API_KEY=...

What the skill handles

Same SKILL.md format in Claude Code and Codex CLI, so one install covers both environments.

Two-step upload

Request a pre-signed URL, PUT the file, then submit for processing. The skill handles the branching and retries.

Polling with backoff

Polls GET /v1/jobs/{id} until the job settles, backing off on 429s and never busy-looping.

Webhook signature verification

HMAC-SHA256 over timestamp and body. The skill generates the verification code in whatever language your project uses.

Structured extraction schemas

Pass a JSON schema up to 64 KB and get typed data back, with field-level validation. No templates to build first.

Correct failure semantics

An HTTP 200 with job.status "failed" is a job failure, not a success. The skill teaches the agent to tell the difference.

Keys in env vars only

The skill reads OCRWELL_API_KEY from the environment. Generated code never hard-codes secrets or commits them.

Full install and verification steps

Cost Comparison

Save with OCRWell

Using our dedicated OCR processing leads to better accuracy and big savings vs using in-built image processing.

Per A4 page	OCRWell Scale + Opus 4.7 text	Raw 200 DPI image → Opus 4.7
OCR cost	$0.0008 ($600 ÷ 750,000 pages)	—
Input to Claude	~700 text tokens	~4,784 image tokens
Opus 4.7 input ($5 / MTok)	$0.0035	$0.0239
Total per page	$0.0043	$0.0239

Opus 4.7 API pricing from Anthropic; image token count uses the published width × height ÷ 750 formula, capped at Opus 4.7's 4,784-token native resolution limit. Output tokens are identical in both paths and excluded. Scale Plan per-page OCR cost assumes 100% utilisation of the 750,000 included pages and no overage.

Ship the document-reading part of your app this afternoon.

Generate a key, install the skill, and your next prompt can turn real-world paperwork into data your code can work with.

Get your API key No credit card required. 200 OCR + 50 extraction pages / month, free forever.