PyMuPDF

PyMuPDF

The PDF engine behind over 50 million monthly downloads, powering AI pipelines worldwide.

PyMuPDF is a high-performance Python library for data extraction, layout analysis, conversion, rendering and manipulation of PDF (and other) documents. Built on top of MuPDF — a lightweight, fast C engine — PyMuPDF gives you precise, low-level control over documents alongside high-level convenience APIs.

---

Why PyMuPDF?

• Fast — powered by MuPDF, a best-in-class C rendering engine
• Accurate — pixel-perfect text extraction with font, color, and position metadata
• Versatile — read, write, annotate, redact, merge, split, and convert documents
• LLM-ready — structured data with JSON & native Markdown output for RAG and AI pipelines
• Grounded results — improved, verifiable results, see: PyMuPDF & Grounding

---

Installation

pip install --upgrade pymupdf

If you have a broken installation due to upgrading PyMuPDF from earlier versions, or from running separate installation commands for pymupdf-layout or pymupdf4llm, then please use:

pip install --force-reinstall --no-cache-dir pymupdf

This will ensure a clean installation of PyMuPDF.

Wheels

Wheels are available for Windows, macOS, and Linux on Python 3.11–3.15. If no pre-built wheel exists for your platform, pip will compile from source (requires a C/C++ toolchain).

PyMuPDF uses the ONNX Runtime for certain layout analysis features - ensure your Python environment can support it. For more details see PyMuPDF Layout & ONNX Runtime

PyMuPDF Config

After installation, run the pymupdf-config command, This will review what is installed and show information about what other packages (Python and otherwise) may be available for your system and could improve accuracy and speed.

Optional extras

Package	Purpose
pymupdf-office	Adds Office document support
pymupdf-fonts	Extended font collection for text output
rapidocr	OCR engine (recommended)
tesseract-ocr	OCR engine

# Office support
pip install pymupdf-office

# More fonts
pip install pymupdf-fonts

# OCR
pip install rapidocr

# OCR (Tesseract must be installed separately)
# macOS
brew install tesseract

# Ubuntu / Debian
sudo apt install tesseract-ocr

---

OCR Support

We compared both OCR engines on the ExtractBench document corpus:

Document length	PDFs	PyMuPDF + Tesseract	PyMuPDF + RapidOCR
Short, 1–10 pages	252	80.39%	88.94%
Medium, 11–50 pages	98	82.72%	87.00%
Long, over 50 pages	20	30.49%	29.22%
Overall	370	78.31%	85.20%

Overall, for the best OCR accuracy, it is recommended to use RapidOCR.

Supported File Formats

Input

Category	Formats
PDF & derivatives	PDF, XPS, EPUB, CBZ, MOBI, FB2, SVG, TXT, MD
Images	PNG, JPEG, BMP, TIFF, GIF, and more
PyMuPDF Office	DOC, DOCX, XLS, XLSX, PPT, PPTX, HWP, HWPX

Output

Format	Notes
PDF	Full fidelity conversion from Office formats
SVG	Vector page rendering
Image (PNG, JPEG, …)	Page rasterisation at any DPI
Markdown	Structure-aware, LLM-ready
JSON	Bounding boxes, layout data, per-element detail
Plain text	Fast, lightweight extraction

---

Quick start

Extract text

import pymupdf

doc = pymupdf.open("document.pdf")
for page in doc:
    print(page.get_text())

Extract text with layout metadata

import pymupdf

doc = pymupdf.open("document.pdf")
page = doc[0]

blocks = page.get_text("dict")["blocks"]
for block in blocks:
    if block["type"] == 0:  # text block
        for line in block["lines"]:
            for span in line["spans"]:
                print(f"{span['text']!r}  font={span['font']}  size={span['size']:.1f}")

Extract tables

import pymupdf

doc = pymupdf.open("spreadsheet.pdf")
page = doc[0]

tables = page.find_tables()
for table in tables:
    print(table.to_markdown())

    # or get as Pandas DataFrame
    df = table.to_pandas()

Render a page to an image

import pymupdf

doc = pymupdf.open("document.pdf")
page = doc[0]

pixmap = page.get_pixmap(dpi=150)
pixmap.save("page_0.png")

OCR a scanned document

import pymupdf

doc = pymupdf.open("scanned.pdf")
page = doc[0]

# Requires Tesseract installed and on PATH
text = page.get_textpage_ocr(language="eng").extractText()
print(text)

Convert Markdown to PDF

import pymupdf

md_doc = pymupdf.open("example.md")
md_doc.save("example.pdf")

Convert to Markdown for LLMs

import pymupdf

doc = pymupdf.open("report.pdf")
md = doc.to_markdown()
# Pass directly to your LLM or vector store
print(md)

Annotate and redact

import pymupdf

doc = pymupdf.open("contract.pdf")
page = doc[0]

# Add a highlight annotation
rect = pymupdf.Rect(72, 100, 400, 120)
page.add_highlight_annot(rect)

# Add a redaction and apply it
page.add_redact_annot(rect)
page.apply_redactions()

doc.save("contract_redacted.pdf")

Merge PDFs

import pymupdf

merger = pymupdf.open()
for path in ["part1.pdf", "part2.pdf", "part3.pdf"]:
    merger.insert_pdf(pymupdf.open(path))

merger.save("merged.pdf")

Convert an Office document to PDF

import pymupdf.office

pymupdf.office.unlock("YOUR-LICENSE-KEY")

doc = pymupdf.open("presentation.pptx")
pdf_bytes = doc.convert_to_pdf()

with open("output.pdf", "wb") as f:
    f.write(pdf_bytes)

Extract LLM-ready Markdown from a Word document

import pymupdf.office

pymupdf.office.unlock("YOUR-LICENSE-KEY")

pymupdf.office.to_markdown("input.docx", "output.md")

---

Features

Core capabilities

Feature	Description
Text extraction	Plain text, rich dict (font, size, color, bbox), HTML, XML, raw blocks
Table detection	find_tables() — locate, extract, and export tables as Markdown or structured data
Image extraction	Extract embedded images and render any page to a high-resolution Pixmap
Rendering	Render PDF pages to images or Pixmap data for use in UI or other workflows
OCR	Tesseract integration — full-page or partial OCR, configurable language
Annotations	Read and write highlights, underlines, squiggly lines, sticky notes, free text, ink, stamps
Redaction	Add and permanently apply redaction annotations
Forms	Read and fill PDF AcroForm fields
PDF creation	Create PDFs directly with the API or quickly convert from Markdown files
PDF editing	Insert, delete, and reorder pages; set metadata; merge and split documents
Drawing	Draw lines, curves, rectangles, and circles; insert HTML boxes
Encryption	Open password-protected PDFs; save with RC4 or AES encryption
Links	Extract hyperlinks, internal cross-references, and URI targets
Bookmarks	Read and write the outline / table of contents tree
Metadata	Title, author, creation date, producer, subject, and custom entries
Color spaces	RGB, CMYK, greyscale; color space conversion

LLM & AI output

doc = pymupdf.open("document.pdf") , then:

Output	API
Markdown	doc.to_markdown(path)
JSON	doc.to_json(path)
Plain text	doc.to_text(path)

Supports multi-column layouts, natural reading order and page chunking.

---

Supported Python versions

Python 3.11 – 3.15 (as of v2.0). Wheels ship for:

• manylinux x86_64 and aarch64
• musllinux x86_64
• macOS x86_64 and arm64
• Windows x86 and x86_64

---

Performance

PyMuPDF is built on MuPDF — one of the fastest PDF rendering engines available. Typical benchmarks against pure-Python PDF libraries show 10–50× speed improvements for text extraction and 100× or more for page rendering, with a minimal memory footprint.

PyMuPDF processes documents without a GPU, cutting infrastructure costs significantly compared to vision-based LLM approaches.

---

Recipes

import pymupdf
from pathlib import Path

doc = pymupdf.open("document.pdf")
out = Path("images")
out.mkdir(exist_ok=True)

for page_index, page in enumerate(doc):
    for img_index, img in enumerate(page.get_images()):
        xref = img[0]
        pix = pymupdf.Pixmap(doc, xref)
        if pix.n > 4:  # convert CMYK
            pix = pymupdf.Pixmap(pymupdf.csRGB, pix)
        pix.save(out / f"page{page_index}_img{img_index}.png")

import pymupdf

doc = pymupdf.open("document.pdf")
needle = "confidential"

for page in doc:
    hits = page.search_for(needle)
    if hits:
        print(f"Page {page.number}: {len(hits)} occurrence(s)")
        for rect in hits:
            page.add_highlight_annot(rect)

doc.save("highlighted.pdf")

import pymupdf

doc = pymupdf.open("document.pdf")
for i, page in enumerate(doc):
    out = pymupdf.open()
    out.insert_pdf(doc, from_page=i, to_page=i)
    out.save(f"page_{i + 1}.pdf")

import pymupdf

doc = pymupdf.open("document.pdf")
for page in doc:
    page.insert_text(
        point=pymupdf.Point(72, page.rect.height / 2),
        text="DRAFT",
        fontsize=72,
        color=(0.8, 0.8, 0.8),
        rotate=45,
    )

doc.save("watermarked.pdf")

---

Office Document Processing

PyMuPDF can be extended with PyMuPDF Office. This adds a conversion layer that handles Microsoft and Korean Office formats natively — no Office installation, no COM interop, no LibreOffice subprocess.

Once unlocked, pymupdf.open() accepts Office files exactly like PDFs:

import pymupdf.office
pymupdf.office.unlock("YOUR-LICENSE-KEY")

# Works identically regardless of format
for fmt in ["contract.docx", "data.xlsx", "deck.pptx", "report.hwpx"]:
    doc = pymupdf.open(fmt)
    for page in doc:
        print(page.get_text())

Get a trial license key for PyMuPDF Office

What you can do with Office documents:

• Extract text and images page-by-page
• Convert to PDF with doc.convert_to_pdf()
• Rasterise pages to PNG/JPEG for visual inspection

Restrictions Without a License Key

When pymupdf.office.unlock() is called without a key, the following restrictions apply:

Restriction	Detail
Page limit	Only the first 3 pages of any document are accessible
Time limit	Evaluation period — functionality expires after a set duration

All other features work normally within these constraints, making it straightforward to prototype before purchasing a license.

---

Frequently Asked Questions

Can I use PyMuPDF and PyMuPDF Office without sending data to the cloud?

Yes, absolutely — and this is one of PyMuPDF's most significant advantages.

PyMuPDF runs entirely locally. It is a native Python library built on top of the MuPDF C engine. When you call pymupdf.open(), page.get_text(), page.find_tables(), or any other method, everything executes in-process on your own machine. No data is transmitted anywhere.

There are no telemetry calls, no licence validation callbacks, no cloud dependencies of any kind in the open-source AGPL build or the commercial build. Once the package is installed, it works fully air-gapped.

This makes PyMuPDF well-suited for:

• Regulated industries — healthcare (HIPAA), finance, legal, government, where documents cannot leave a controlled environment
• On-premise deployments — servers with no outbound internet access
• Air-gapped systems — classified or sensitive environments
• Self-hosted RAG pipelines — processing confidential documents locally before feeding an on-premise LLM
• Saving on token costs for document pre-processing before sending data to your LLM

The only thing you need an internet connection for is the initial pip install. After that, the package and all its capabilities are entirely self-contained.

Should I import pymupdf or import fitz?

Use import pymupdf. The fitz name is a legacy alias that still works as of v1.24.0+, but import pymupdf is the recommended and future-proof approach. The two are interchangeable in existing code:

import pymupdf          # recommended
# import fitz           # legacy alias — still works but avoid for new code

Does PyMuPDF work with Korean, Japanese, or Chinese documents?

Yes — PyMuPDF has solid CJK support

How do I extract Markdown from PDF for LLM?

This is the best starting point for feeding an LLM or building a RAG pipeline.

import pymupdf

doc = pymupdf.open("report.pdf")
md = doc.to_markdown()
print(md)
# Tables appear as Markdown | col1 | col2 | ... inline with the text

Text extraction returns garbled characters or empty output. Why?

This usually means the PDF uses custom font encodings without a proper character map (CMAP). The font's glyphs are present but cannot be mapped back to Unicode. In these cases:

• Use OCR as a fallback (page.get_textpage_ocr())
• Consider that scanned PDFs will always need OCR — text extraction on scans returns nothing

How do I extract text from a specific area of a page?

Pass a clip rectangle to get_text():

import pymupdf

doc = pymupdf.open("input.pdf")
page = doc[0]

# Define the area you want (x0, y0, x1, y1) in points
clip = pymupdf.Rect(50, 100, 400, 300)
text = page.get_text("text", clip=clip)

How do I search for text and find its location on the page?

import pymupdf

doc = pymupdf.open("input.pdf")
page = doc[0]

# Returns a list of Rect objects surrounding each match
locations = page.search_for("invoice number")
for rect in locations:
    print(rect)  # e.g. Rect(72.0, 120.5, 210.0, 134.0)

get_images shows no images but I can clearly see charts in the PDF. Why?

Charts and diagrams created by tools like matplotlib, Excel, or R are typically rendered as vector graphics (PDF drawing commands), not raster images. get_images only lists embedded raster image objects and will not detect vector graphics. To capture these, rasterise the entire page with page.get_pixmap().

How does OCR work in PyMuPDF? Does it require a separate Tesseract installation?

PyMuPDF uses MuPDF's built-in Tesseract-based OCR support, so there is no Python-level pytesseract dependency. However, PyMuPDF still needs access to the Tesseract language data files (tessdata), and automatic tessdata discovery may invoke the tesseract executable (for example, to list available languages) if you do not explicitly provide a tessdata path. In practice, the recommended setup is to either install Tesseract so discovery works automatically, or configure the tessdata location yourself via the tessdata parameter or the TESSDATA_PREFIX environment variable. Over 100 languages are supported.

PyMuPDF also supports RapidOCR - which is the recommended OCR engine - RapidOCR is a pip-installable Python package that runs ONNX models, so it doesn't need a separate Tesseract install or TESSDATA_PREFIX.

import pymupdf

doc = pymupdf.open("scanned.pdf")
page = doc[0]

# Get a text page using OCR
tp = page.get_textpage_ocr(language="eng")
text = page.get_text(textpage=tp)
print(text)

How do I run OCR on a standalone image file (not a PDF)?

import pymupdf

pix = pymupdf.Pixmap("image.png")
if pix.alpha:
    pix = pymupdf.Pixmap(pix, 0)  # remove alpha channel — required for OCR

# Wrap in a 1-page PDF and OCR it
doc = pymupdf.open()
page = doc.new_page(width=pix.width, height=pix.height)
page.insert_image(page.rect, pixmap=pix)
tp = page.get_textpage_ocr()
text = page.get_text(textpage=tp)

How do I highlight text in a PDF?

import pymupdf

doc = pymupdf.open("input.pdf")
page = doc[0]

# Use quads=True for accurate highlights on non-horizontal text
quads = page.search_for("important term", quads=True)
page.add_highlight_annot(quads)

doc.save("highlighted.pdf")

PyMuPDF supports all standard PDF text markers: highlight, underline, strikeout, and squiggly.

How do I permanently redact (remove) content from a PDF?

Redaction is a deliberate two-step process so you can review before committing:

import pymupdf

doc = pymupdf.open("input.pdf")
page = doc[0]

# Step 1: Mark the area(s) to redact
rect = page.search_for("confidential")[0]
page.add_redact_annot(rect, fill=(1, 1, 1))  # white fill

# Step 2: Apply — permanently removes the underlying content
page.apply_redactions()

doc.save("redacted.pdf")

After apply_redactions(), the original content is gone. It cannot be recovered from the saved file.

How do I read form field values from a PDF?

import pymupdf

doc = pymupdf.open("form.pdf")
page = doc[0]

for field in page.widgets():
    print(f"{field.field_name}: {field.field_value}")

How do I fill in a PDF form programmatically?

import pymupdf

doc = pymupdf.open("form.pdf")
page = doc[0]

for field in page.widgets():
    if field.field_name == "First Name":
        field.field_value = "Ada"
        field.update()

doc.save("filled_form.pdf")

Can I use multithreading with PyMuPDF?

No. PyMuPDF does not support multithreaded use, even with Python's newer free-threading mode. The underlying MuPDF library only provides partial thread safety, and a fully thread-safe PyMuPDF implementation would still impose a single-threaded overhead — negating the benefit.

Use multiprocessing instead. Each process opens the file independently and works on its own page range:

from multiprocessing import Pool
import pymupdf

def process_pages(args):
    path, start, end = args
    doc = pymupdf.open(path)  # each process opens its own handle
    results = []
    for i in range(start, end):
        results.append(doc[i].get_text())
    return results

with Pool(4) as pool:
    chunks = [("input.pdf", 0, 25), ("input.pdf", 25, 50), ...]
    all_results = pool.map(process_pages, chunks)

How can I speed up repeated text extraction on the same page?

Reuse a TextPage object. Creating a TextPage is the expensive part — once created, switching between extraction formats is cheap:

import pymupdf

page = doc[0]
tp = page.get_textpage()  # create once

text  = page.get_text("text",    textpage=tp)
words = page.get_text("words",   textpage=tp)
data  = page.get_text("dict",    textpage=tp)

This can reduce execution time by 50–95% for repeated extractions on the same page.

How do I read and write PDF metadata?

import pymupdf

doc = pymupdf.open("input.pdf")

# Read
print(doc.metadata)
# {'title': '...', 'author': '...', 'subject': '...', 'keywords': '...', ...}

# Write
doc.set_metadata({
    "title": "Annual Report 2025",
    "author": "Finance Team",
    "keywords": "annual, finance, 2025"
})
doc.save("output.pdf")

How do I read or set the table of contents / bookmarks?

import pymupdf

doc = pymupdf.open("input.pdf")

# Read — returns a list of [level, title, page_number] entries
toc = doc.get_toc()
for level, title, page in toc:
    print(" " * level, title, "→ page", page)

# Write
new_toc = [
    [1, "Introduction",  1],
    [1, "Methods",       5],
    [2, "Data sources",  6],
]
doc.set_toc(new_toc)
doc.save("output.pdf")

---

Documentation

Full installation guide, API reference, cookbook, and tutorial at pymupdf.readthedocs.io.

• Installation guide
• API reference
• Cookbook
• Tutorial
• Changelog
• PyMuPDF Office docs

---

Licensing

PyMuPDF and MuPDF are maintained by Artifex Software, Inc.

• Open source — GNU AGPL v3. Free for open-source projects.
• Commercial — separate commercial licences available from Artifex for proprietary applications.

---

Contributing

Contributions are welcome. Please open an issue before submitting large pull requests.

• Issue tracker
• Discord community

⭐ Support this project

If you find this useful, please consider giving it a star — it helps others discover it!