Pypdf Migration Check — find imports and show replacement lines

Migrate from PyMuPDF to pypdf: 6 findings in one invoice worker

6 findings in a 13-line invoice export worker: that is what the checker printed for a Python backend team running PyMuPDF behind a customer portal.

What Artifex states

Artifex licenses PyMuPDF under the GNU AGPLv3 or a commercial licence. Its licensing page (artifex.com/licensing) says you cannot deploy the open-source edition as part of a server-based application or service without disclosing your own application's full source code under AGPL to any users interacting with it, and that if you cannot meet the AGPL requirements a commercial licence is required. There is no public price on that page. The exposure starts the day the service is deployed to users.

pymupdf4llm, the LLM and RAG helper, is published on PyPI under the same dual licence. So is the binary wheel PyMuPDFb.

What the sample file looked like

The worker opened a PDF, joined the text of every page, rendered page one to a PNG thumbnail, searched the last page for "Total due" and pulled its tables. Six lines touch PyMuPDF:

import fitz -> from pypdf import PdfReader, PdfWriter fitz.open(path) -> reader = PdfReader(path) page.get_text() -> page.extract_text() page.get_pixmap(dpi=110) -> pypdfium2.PdfDocument(path)[i].render(scale=dpi/72).to_pil() page.search_for("Total due") -> pdfplumber.open(path).pages[i].search("text") page.find_tables() -> pdfplumber.open(path).pages[i].extract_tables()

After the rewrite the same file returns 0 findings. pypdf (BSD-3-Clause) handles reading, text, merging and metadata. pdfplumber (MIT) handles word positions, search and tables. pypdfium2 (BSD-3-Clause / Apache-2.0) renders pages to images. The versions in the fixes were read on PyPI on 2026-09-30: pypdf 6.19.0, pdfplumber 0.11.10, pypdfium2 5.13.0.

Why a package scanner is not enough

pip-licenses and similar tools tell you that pymupdf is AGPL. They stop there. They do not find the six call sites, and they cannot tell you that get_pixmap needs a different library from get_text. Chatbots often answer with PyPDF2, which the project itself retired in favour of pypdf. And the PyPI package literally named fitz is an unrelated 0.0.0 project, so a requirements line reading fitz is a bug on top of a licence question.

The 25 rules

Five dependency rules cover requirements*.txt, pyproject.toml (PEP 621 and Poetry) and conda environment.yml: pymupdf, PyMuPDFb, pymupdf4llm, pymupdfpro and fitz. Six import rules cover import fitz, import pymupdf, from fitz import, pymupdf4llm, the LangChain PyMuPDFLoader and the LlamaIndex PyMuPDFReader. Fourteen call rules cover open, Matrix and Rect, get_text, get_pixmap, search_for, find_tables, insert_pdf, page_count, get_toc, set_metadata, set_rotation, get_images, insert_text and authenticate. Call rules only fire in a file that imports fitz or pymupdf, so a Tkinter get_text call stays quiet.

How to run it

Open the file in VS Code with the extension installed, or paste it into the web version. The Problems panel lists each line with its replacement.

What comes next

A single file is where most people start. The next job is the whole repository: every service, every requirements file, and proof for the company that nobody adds import fitz back next quarter. That is the workspace plan, the dated exposure report and the CI gate.

This is not legal advice. Read the Artifex terms with your counsel before deciding between a commercial licence and a migration.