Installation - PyMuPDF documentation (original) (raw)

Requirements¶

All the examples below assume that you are running inside a Python virtual environment. See: https://docs.python.org/3/library/venv.html for details. We also assume that pip is up to date.

For example:

• Windows:
  py -m venv pymupdf-venv
  .\pymupdf-venv\Scripts\activate
  python -m pip install --upgrade pip
• Linux, MacOS:
  python -m venv pymupdf-venv
  . pymupdf-venv/bin/activate
  python -m pip install --upgrade pip

Installation¶

PyMuPDF should be installed using pip with:

pip install --upgrade pymupdf

This will install from a Python wheel if one is available for your platform.

Installation when a suitable wheel is not available¶

If a suitable Python wheel is not available, pip will automatically build from source using a Python sdist.

This requires C/C++ development tools to be installed:

• On Windows:
  • Install Visual Studio 2019. If not installed in a standard location, set environmental variable PYMUPDF_SETUP_DEVENV to the location of thedevenv.com binary.
  • Having other installed versions of Visual Studio, for example Visual Studio 2022, can cause problems because one can end up with MuPDF and PyMuPDF code being compiled with different compiler versions.

The build will automatically download and build MuPDF.

Problems after installation¶

• On Windows, Python error:
  ImportError: DLL load failed while importing _extra
  This has been occasionally seen if MSVCP140.dll is missing, and appears to be caused by a bug in some versions (2015-2017) of Microsoft Visual C++ Redistributables.
  It is recommended to search for MSVCP140.dll in https://msdn.comto find instructions for how to reinstall it. For examplehttps://learn.microsoft.com/cpp/windows/latest-supported-vc-redist has permalinks to the latest supported versions.
  See https://github.com/pymupdf/PyMuPDF/issues/2678 for more details.
• Python error:
  ModuleNotFoundError: No module named 'frontend'
  This can happen if PyMuPDF’s legacy name fitz is used (for example import fitz instead of import pymupdf), and an unrelated Python package calledfitz (https://pypi.org/project/fitz/) is installed.
  The fitz package appears to be no longer maintained (the latest release is from 2017), but unfortunately it does not seem possible to remove it from pypi.org. It does not even work on its own, as well as breaking the use of PyMuPDF’s legacy name.
  There are a few ways to avoid this problem:
  • Use import pymupdf instead of import fitz, and update one’s code to match.
  • Or uninstall the fitz package and reinstall PyMuPDF:
    pip uninstall fitz
    pip install --force-reinstall pymupdf
  • Or use import pymupdf as fitz. However this has not been well tested.
• With Jupyter labs on Apple Silicon (arm64), Python error:
  ImportError: /opt/conda/lib/python3.11/site-packages/pymupdf/libmupdf.so.24.4: undefined symbol: fz_pclm_write_options_usage
  This appears to be a problem in Jupyter labs; see:https://github.com/pymupdf/PyMuPDF/issues/3643#issuecomment-2210588778.
• On Windows, Python error:
  ImportError: dynamic module does not define module export function (PyInit__extra)
  This was reported 2025-03-26 in https://github.com/pymupdf/PyMuPDF/issues/4405.
  The fix appears to be to install the latest VC_redist.x64.exe.

Notes¶

• Wheels are available for the following platforms:

  • Windows 32-bit Intel.
  • Windows 64-bit Intel.
  • Linux 64-bit Intel.
  • Linux 64-bit ARM.
  • MacOS 64-bit Intel.
  • MacOS 64-bit ARM.
    Details:

  • We release a single wheel for each of the above platforms.
  • Each wheel uses the Python Stable ABI of the current oldest supported Python version (currently 3.10), and so works with all later Python versions, including new Python releases.
  • Wheels are tested on all Python versions currently marked as “Supported” on https://devguide.python.org/versions/, currently 3.9, 3.10, 3.11, 3.12, 3.13 and 3.14.
• Wheels are not available for Python installed with Chocolatey on Windows. Instead install Python using the Windows installer from the python.org website, see:http://www.python.org/downloads
• Wheels are not available for Linux-aarch64 with Musl libc (For example Alpine Linux on aarch64), and building from source is known to fail.
• There are no mandatory external dependencies. However, some optional feature are available only if additional components are installed:
  • Pillow is required for Pixmap.pil_save() and Pixmap.pil_tobytes().
  • fontTools is required for Document.subset_fonts().
  • pymupdf-fonts is a collection of nice fonts to be used for text output methods.
  • Tesseract-OCR for optical character recognition in images and document pages. Tesseract is separate software, not a Python package. To enable OCR functions in PyMuPDF, Tesseract must be installed and the tessdata folder name specified; see below.
    Note
    You can install these additional components at any time – before or after installing PyMuPDF. PyMuPDF will detect their presence during import or when the respective functions are being used.

Build and install from a local PyMuPDF source tree¶

Initial setup:

• Install C/C++ development tools as described above.
• Enter a Python venv and update pip, as described above.
• Get a PyMuPDF source tree:
  • Clone the PyMuPDF git repository:
    git clone https://github.com/pymupdf/PyMuPDF.git
  • Or download and extract a .zip or .tar.gz source release fromhttps://github.com/pymupdf/PyMuPDF/releases.

Then one can build PyMuPDF in two ways:

• Build and install PyMuPDF with default MuPDF version:
  cd PyMuPDF && pip install .
  This will automatically download a specific hard-coded MuPDF source release, and build it into PyMuPDF.
• Or build and install PyMuPDF using a local MuPDF source tree:
  • Clone the MuPDF git repository:
    git clone --recursive https://git.ghostscript.com/mupdf.git
  • Build PyMuPDF, specifying the location of the local MuPDF tree with the environmental variables PYMUPDF_SETUP_MUPDF_BUILD:
    cd PyMuPDF && PYMUPDF_SETUP_MUPDF_BUILD=../mupdf pip install .

Also, one can build for different Python versions in the same PyMuPDF tree:

• PyMuPDF will build for the version of Python that is being used to runpip. To run pip with a specific Python version, use python -m pipinstead of pip.
  So for example on Windows one can build different versions with:
  cd PyMuPDF && py -3.10 -m pip install .
  or:
  cd PyMuPDF && py -3.10-32 -m pip install .

Running tests¶

Having a PyMuPDF tree available allows one to run PyMuPDF’s pytest test suite:

pip install pytest fontTools pytest PyMuPDF/tests

Notes about using a non-default MuPDF¶

Using a non-default build of MuPDF by setting environmental variablePYMUPDF_SETUP_MUPDF_BUILD can cause various things to go wrong and so is not generally supported:

• If MuPDF’s major version number differs from what PyMuPDF uses by default, PyMuPDF can fail to build, because MuPDF’s API can change between major versions.
• Runtime behaviour of PyMuPDF can change because MuPDF’s runtime behaviour changes between different minor releases. This can also break some PyMuPDF tests.
• If MuPDF was built with its default config instead of PyMuPDF’s customised config (for example if MuPDF is a system install), it is possible thattests/test_textbox.py:test_textbox3() will fail. One can skip this particular test by adding -k 'not test_textbox3' to the pytestcommand line.

Official PyMuPDF Linux wheels may not install on older Linux systems¶

Releases of PyMuPDF are incompatible with older Linux systems.

For example as of 2025-09-03, pip install pymupdf does not work on some AWS Lambda systems - see https://github.com/pymupdf/PyMuPDF/discussions/4631.

This is because official PyMuPDF Linux wheels are built with a version of glibc determined by the current Python manylinux environment. These wheels are incompatible with Linux systems that have an older glibc.

The official Python manylinux environment is updated periodically to use newer glibc versions, so new releases of PyMuPDF become increasingly incompatible with older Linux systems.

There is nothing that can be done about this, other than updating older Linux systems, or building PyMuPDF locally from source.

For more details, please see: Python Packaging Authority.

Packaging¶

See Packaging for Linux distributions.

Using with Pyodide¶

See Pyodide.

Enabling Integrated OCR Support¶

PyMuPDF will already contain all the logic to support OCR functions. But it additionally does need Tesseract’s language support data.

If not specified explicitly, PyMuPDF will attempt to find the installed Tesseract’s tessdata, but this should probably not be relied upon.

Otherwise PyMuPDF requires that Tesseract’s language support folder is specified explicitly either in PyMuPDF OCR functions’ tessdata arguments oros.environ["TESSDATA_PREFIX"].

So for a working OCR functionality, make sure to complete this checklist:

1. Locate Tesseract’s language support folder. Typically you will find it here:
   • Windows: C:/Program Files/Tesseract-OCR/tessdata
   • Unix systems: /usr/share/tesseract-ocr/4.00/tessdata
2. Specify the language support folder when calling PyMuPDF OCR functions:
   • Set the tessdata argument.
   • Or set os.environ["TESSDATA_PREFIX"] from within Python.
   • Or set environment variable TESSDATA_PREFIX before running Python, for example:
     * Windows: setx TESSDATA_PREFIX "C:/Program Files/Tesseract-OCR/tessdata"
     * Unix systems: declare -x TESSDATA_PREFIX=/usr/share/tesseract-ocr/4.00/tessdata

Note

English language support is included by default in Tesseract installation.

Tesseract Language Packs for other languages must be installed separately, and the tessdata folder must be specified to PyMuPDF as described above, for OCR to work with those languages.