mkdocs-jupyter: Use Jupyter Notebooks in mkdocs

• Docs demo Site
• Add Jupyter Notebooks directly to the mkdocs navigation
• Support for multiple formats:
  • .ipynb and .py files (using jupytext)
• Same style as regular Jupyter Notebooks
  • Support Jupyter Themes
• Option to execute the notebook before converting
• Support for ipywidgets
• Support for mkdocs TOC
• Option to include notebook source

Installation

pip install mkdocs-jupyter

Configuration

In the mkdocs.yml use Jupyter notebooks (.ipynb) or Python scripts (.py) as pages:

nav:
    - Home: index.md
    - Notebook page: notebook.ipynb
    - Python file: python_script.py
plugins:
    - mkdocs-jupyter

Markdown cell processor

Markdown cells use Jupyter's nbconvert renderer by default (markdown_cell_processor: jupyter). To process them with MkDocs and its configured Markdown extensions, opt in with:

plugins:
    - mkdocs-jupyter:
          markdown_cell_processor: mkdocs
    - mkdocstrings

With mkdocstrings installed and configured, notebook Markdown cells can then reference documented API objects, for example [Model.fit][my_package.Model.fit]. Code cells and rich outputs still use nbconvert.

In this mode, Markdown syntax, heading IDs and the table of contents follow MkDocs and your markdown_extensions. For notebook math, configure pymdownx.arithmatex with generic: true; Mermaid fences likewise need your site's configured Mermaid extension, such as Material's SuperFences setup. References and Markdown extensions are processed on every build, including when notebook conversion is cached.

Titles and Table of Contents

The first h1 header (#) in your notebook will be used as the title.

# This H1 header will be the title.

This can be turned off in the configuration (in which case the filename will be used as title):

plugins:
    - mkdocs-jupyter:
          ignore_h1_titles: True

In order to see the table of contents you need to maintain a hierarchical headers structure in your notebooks. You must use h2 headers (##) and not h1 (#)

## This H2 title will show in the table of contents

If you want to nest headers in the TOC you need to add additional levels later in the same markdown cell or new bottom markdown cells:

## This header will show as top level in the table of contents

<content>

### This one will be displayed inside the above level

Notebook page metadata

Add YAML frontmatter at the start of the notebook's first Markdown cell to set standard MkDocs page metadata:

---
title: Analysis
description: An overview of the analysis.
hide:
  - navigation
  - toc
search:
  exclude: true
---

# Analysis

Notebook content goes here.

The frontmatter is removed from the rendered notebook and exposed as page.meta to MkDocs plugins and templates. In Material, the example hides the page's navigation and table of contents and excludes it from search. Other metadata keys are passed through unchanged.

Search exclusion uses Material's search plugin. With older MkDocs versions, select it explicitly as material/search in your plugins configuration.

The first Markdown cell can contain only frontmatter, with content in later cells. Frontmatter must be a closed YAML mapping; YAML examples in code fences or later Markdown cells remain content. This also works in Jupytext notebooks.

For notebooks with frontmatter, an explicit nav title takes precedence over the metadata title, followed by the first H1. Set ignore_h1_titles: true to ignore that H1 fallback. Themes may use the metadata title separately, such as for the browser tab title. Notebooks without frontmatter keep their existing title behavior.

Including or Ignoring Files

You can control which files are included or ignored via lists of glob patterns:

plugins:
    - mkdocs-jupyter:
          include: ["*.ipynb"] # Default: ["*.py", "*.ipynb"]
          ignore: ["some-irrelevant-files/*.ipynb"]

Notebooks outside the docs directory

Use notebook_dirs to mount source directories outside docs_dir:

plugins:
    - mkdocs-jupyter:
          notebook_dirs:
              examples: notebooks

nav:
    - Home: index.md
    - Example: examples/tutorial.ipynb

Execute Notebook

You can tell the plugin to execute the notebook before converting, default is False:

plugins:
    - mkdocs-jupyter:
          execute: true

You can tell the plugin to ignore the execution of some files (with glob matching):

plugins:
    - mkdocs-jupyter:
          execute_ignore:
              - "my-secret-files/*.ipynb"

To fail when notebook execution fails set allow_errors to false:

plugins:
    - mkdocs-jupyter:
          execute: true
          allow_errors: false

Kernel

By default the plugin will use the kernel specified in the notebook to execute it. You can specify a custom kernel name to use for all the notebooks:

plugins:
    - mkdocs-jupyter:
          kernel_name: python3

Ignore Code Input

By default the plugin will show full code and regular cell output details. You can hide cell code input for all the notebooks:

plugins:
    - mkdocs-jupyter:
          show_input: False

You can also decide to hide the Out[#] output notation and other cell metadata for all the notebooks:

plugins:
    - mkdocs-jupyter:
          no_input: True

You can also only hide the In[#] and Out[#] prompts but keep the input and output content:

plugins:
    - mkdocs-jupyter:
          no_prompt: True

Remove Cell Using Tags

By default the plugin will show full code and regular cell output details. You can hide cell code input for specific cells using tags:

plugins:
    - mkdocs-jupyter:
          remove_tag_config:
              remove_input_tags:
                  - hide_code

More detailed on removing cell based on tag, see NbConvert Customization)

Jupyter themes

You can configure the different Jupyter themes. For example if using material with slate color scheme you can use the Jupyter Lab dark theme:

plugins:
    - mkdocs-jupyter:
          theme: dark

theme:
    name: material
    palette:
        scheme: slate

If a Material-derived theme uses a different name for its dark color scheme, set dark_scheme to that name. The default is slate:

plugins:
    - mkdocs-jupyter:
          dark_scheme: dark
          # Optionally pass the css class for the dark selectors
          dark_selector: ".dark"

Custom notebook templates

Use custom_template to extend the bundled notebook HTML template. Paths are relative to mkdocs.yml; absolute paths also work.

plugins:
  - mkdocs-jupyter:
      custom_template: overrides/notebook.html.j2
      no_prompt: true

For example, overrides/notebook.html.j2 can make cell outputs collapsible:

{% extends "mkdocs_html/notebook.html.j2" %}

{% block output_group %}
<details>
  <summary>Show output</summary>
  {{ super() }}
</details>
{% endblock output_group %}

Includes and local assets are resolved from the custom template's directory before the bundled templates. Notebook heading extraction remains unchanged. Conversion caching is disabled with custom_template so edits to templates, includes, and assets take effect on each build. With execute: true, this also means notebooks execute on each build.

HTML exporter options

Pass additional nbconvert HTML exporter options through html_exporter_options:

plugins:
    - mkdocs-jupyter:
          html_exporter_options:
              embed_images: true
              anchor_link_text: "Link"

embed_images embeds local images referenced in Markdown cells as base64 URLs, including Markdown images and HTML <img> tags. Paths resolve relative to the notebook, so the same references work in Jupyter and the built documentation. Existing image alt text and HTML attributes are retained. Remote images remain external, and missing local images keep their original URLs. Images in arbitrary HTML outputs are outside this nbconvert option's scope.

Option names and value types are validated against nbconvert's configurable HTMLExporter traits. Use plain names such as embed_images, without an HTMLExporter. prefix. Unknown options and invalid values produce a configuration error.

Dedicated plugin settings take precedence: show_input controls input visibility, and no_input / no_prompt keep their prompt-hiding behavior. Use the plugin's custom_template, theme and MathJax settings to configure those features; exporter options do not replace the package's templates, filters or preprocessing pipeline.

Conversion caching remains enabled. Exporter options and the contents of referenced local images are checked before reusing cached HTML. Changing, removing or adding an image triggers reconversion; unchanged images reuse the cache. With execute: true, reconversion also reruns the notebook.

Extra CSS classes

This option will add a custom CSS class to the div container that highlights the code cells. This can be useful to add custom styles to the code cells.

plugins:
  - mkdocs-jupyter:
      highlight_extra_classes: "custom-css-classes"

RequireJS

By default RequireJS is not loaded. This is required for Plotly. You can enable it with:

plugins:
    - mkdocs-jupyter:
          include_requirejs: true

MathJax

Notebook math uses MathJax 4.1.3 by default, including inline equations, display math, AMS numbering and labels. Material's instant navigation is supported. MathJax loads when a notebook is opened; an existing MathJax 3 or 4 runtime is reused, preserving the site's configuration.

To select another supported major version or host MathJax yourself:

plugins:
    - mkdocs-jupyter:
          mathjax_version: 3 # Supported: 2, 3, 4. Default: 4.
          # Optional: URL of the matching MathJax component and its assets.
          custom_mathjax_url: /assets/mathjax/tex-mml-chtml.js

Without custom_mathjax_url (or with an empty value), the plugin uses a CDN URL for the selected version. Version 3 uses MathJax 3.2.2; version 2 keeps the legacy nbconvert configuration. If you already specify a MathJax 2 URL, add mathjax_version: 2 to keep using it. Custom version 3/4 configuration uses window.MathJax and the promise API, rather than MathJax.Hub.

When your site loads MathJax itself, its TeX and processing settings apply to notebooks too. Notebook equations use arithmatex regions with \( / \) and \[ / \] delimiters, which Material's MathJax configuration recognizes. A site that manages rendering should perform one initial typeset: set startup.typeset: false when a navigation callback performs it. MathJax 3 callbacks must serialize asynchronous work through MathJax.startup.promise; MathJax 4 serializes typesetPromise calls itself. The legacy version 2 renderer supports full page loads.

Download notebook link

You can tell the plugin to include the notebook source to make it easy to show a download button in the theme, default is False:

plugins:
    - mkdocs-jupyter:
          include_source: True

This setting will also create a page.nb_url value that you can use in your theme to make a link in each page.

For example in mkdocs-material (see customization), you can create a main.html file like this:

{% extends "base.html" %}

{% block content %}
{% if page.nb_url %}
    <a href="{{ page.nb_url }}" title="Download Notebook" class="md-content__button md-icon">
        {% include ".icons/material/download.svg" %}
    </a>
{% endif %}

{{ super() }}
{% endblock content %}

Caching

By default the plugin caches notebook conversion results so that unchanged notebooks are not re-converted on every build. The cache key is based on the notebook file content and all relevant plugin config options.

To disable caching:

plugins:
    - mkdocs-jupyter:
          cache: false

To change the cache directory (default is .cache/mkdocs-jupyter):

plugins:
    - mkdocs-jupyter:
          cache_dir: .cache/custom-dir

Stale cache entries from previous builds are automatically cleaned up.

Styles

This extensions includes the Jupyter Lab nbconvert CSS styles and does some modifications to make it as generic as possible in order for it to work with a variety of mkdocs themes. This is not always possible and the theme we test the most is mkdocs-material.

It's possible you might need to do some CSS changes to make it look as good as you want, for example for the material theme take a look at their customization docs.

Create a main.html file like:

{% extends "base.html" %}

{% block content %}
{{ super() }}

<style>
// Do whatever changes you need here

.jp-RenderedHTMLCommon p {
    color: red
}

</style>
{% endblock content %}

Mkdocs Material notes

Any markdown specific features such as admonitions won't work with mkdocs-jupyter because those features are not supported by Jupyter itself and we use nbconvert to make the conversion.

To use this type of features you have to define the HTML directly in the markdown cells:

<div class="admonition note">
    <p class="admonition-title">Note</p>
    <p>
        If two distributions are similar, then their entropies are similar,
        implies the KL divergence with respect to two distributions will be
        smaller...
    </p>
</div>