PyPI package

AFIP invoice PDF QR CAE extract and decode

afipcaeqrdecode extracts AFIP CAE invoice metadata from PDF invoices. It renders the first page of the PDF, locates the AFIP QR code, decodes the QR payload, and returns the decoded metadata as a Python dict.

Installation

This package depends on qreader, which depends on pyzbar, which depends on the system zbar library.

On Linux (Ubuntu 22.04):

sudo apt-get install libzbar0

On macOS:

brew install zbar

Then install the package:

pip install afipcaeqrdecode

Quick Start

from afipcaeqrdecode import get_cae_metadata

invoice_metadata = get_cae_metadata("./tests/sample_files/2000005044986390.pdf")

Example output:

{
    "ver": 1,
    "fecha": "2023-02-10",
    "cuit": 30710145764,
    "ptoVta": 4,
    "tipoCmp": 1,
    "nroCmp": 25399,
    "importe": 2460,
    "moneda": "PES",
    "ctz": 1,
    "tipoDocRec": 80,
    "nroDocRec": 30717336905,
    "tipoCodAut": "E",
    "codAut": 73064176949471,
}

Return Value

get_cae_metadata(filepath, attempt_to_repair_json=True) returns:

• a Python dict when AFIP QR metadata is decoded successfully
• None when no AFIP QR metadata can be extracted

The returned dictionary preserves the value types present in the decoded JSON payload. Some invoices encode numeric-looking values as JSON strings, and those values are preserved as strings.

How It Works

The decoding flow has two stages:

1. Render the first page of the PDF with PyMuPDF and run qreader on the resulting image.
2. If that fails, extract embedded images from the first page and run qreader on those images as a fallback.

If the AFIP QR payload contains malformed JSON, json-repair is used by default to repair it before parsing.

Why qreader instead of pyzbar

Earlier versions of this project used only pyzbar. Some real-world AFIP invoice QRs did not decode reliably with pyzbar alone.

qreader still uses pyzbar for decoding, but it first uses a trained QR detector and applies preprocessing strategies that improve decode rates on difficult images.

Notes and Limitations

• On first run, qreader may download model weights before decoding.
• Some invoices omit fields such as fecha in the QR payload.
• This project is tested mainly with sample PDF integration tests in tests/sample_files/.
• This package is still experimental. Use it carefully in production workloads.

Testing

Run the sample-based integration test suite with:

python -m unittest tests.test_sample_files

License

GNU Lesser General Public License v3.0 or later (LGPL-3.0-or-later). See LICENSE.