A Pandoc filter to include figures generated from code blocks using your plotting toolkit of choice.

LTS Haskell 16.31:
Stackage Nightly 2021-02-28:
Latest on Hackage:

See all snapshots pandoc-plot appears in

GPL-2.0-or-later licensed and maintained by Laurent P. René de Cotret
This version can be pinned in stack with:pandoc-plot-,5816

Module documentation for


A Pandoc filter to generate figures from code blocks in documents

Hackage version Stackage version (latest long-term service) Conda Version license

pandoc-plot turns code blocks present in your documents (Markdown, LaTeX, etc.) into embedded figures, using your plotting toolkit of choice, including Matplotlib, ggplot2, MATLAB, Mathematica, and more.


This program is a Pandoc filter. It can therefore be used in the middle of conversion from input format to output format, replacing code blocks with figures.

The filter recognizes code blocks with classes that match plotting toolkits. For example, using the matplotlib toolkit:

# My document

This is a paragraph.

import matplotlib.pyplot as plt

plt.plot([0,1,2,3,4], [1,2,3,4,5])
plt.title('This is an example figure')

Putting the above in, we can then generate the plot and embed it in an HTML page:

pandoc --filter pandoc-plot --output output.html

Note that pandoc-plot only works with pandoc >= 2.11 because of some breaking changes in pandoc’s API.

Supported toolkits

pandoc-plot currently supports the following plotting toolkits (installed separately):

  • matplotlib: plots using the matplotlib Python library;
  • plotly_python : plots using the plotly Python library;
  • plotly_r: plots using the plotly R library
  • matlabplot: plots using MATLAB;
  • mathplot : plots using Mathematica;
  • octaveplot: plots using GNU Octave;
  • ggplot2: plots using ggplot2;
  • gnuplot: plots using gnuplot;
  • graphviz: graphs using Graphviz;
  • bokeh: plots using the Bokeh visualization library;
  • plotsjl: plots using the Julia Plots.jl package.

To know which toolkits are useable on your machine (and which ones are not available), you can check with the toolkits command:

pandoc-plot toolkits

Wish your plotting toolkit of choice was available? Please raise an issue!


You can find more information in the documentation, available either in the source repository file, on the webpage, or via the command pandoc-plot --manual.


Binaries and Installers

Windows, Linux, and Mac OS binaries are available on the GitHub release page. There are also Windows installers.


Like pandoc, pandoc-plot is available as a package installable with conda. Click here to see the package page.

To install in the current environment:

conda install -c conda-forge pandoc-plot


You can install pandoc-plot from the Windows Package Manager winget (just like pandoc). To install:

winget install pandoc-plot

Arch Linux

You can install pandoc-plot from the archlinux user repository as pandoc-plot-bin. You can install using e.g. yay:

yay -S pandoc-plot-bin

From Hackage/Stackage

pandoc-plot is available on Hackage and Stackage. Using the cabal-install tool:

cabal update
cabal install pandoc-plot


stack update
stack install pandoc-plot

From source

Building from source can be done using stack or cabal:

git clone
cd pandoc-plot
stack install # Alternatively, `cabal install`


Change log

pandoc-plot uses Semantic Versioning


  • pandoc-plot will now only render at most N figures in parallel, where N is the number of available CPU cores.
  • Fixed an issue where error message would get mangled in strict-mode.


  • Added the ability to run pandoc-plot in strict mode. By default, pandoc-plot leaves code blocks unchanged if a figure fails to be rendered. In strict mode, pandoc-plot will immediately halt if it encounters a problem, such as a missing toolkit. You can activate strict mode via configuration:
strict: true
  • Added the ability to set command-line arguments for interpreters via configuration. For example, if you want to run the Matplotlib toolkit with all warnings shown:
# Possible parameters for the Matplotlib toolkit
  executable: python
  command_line_arguments: -Wa

Or if you want julia to use more than one thread:

# Possible parameters for the Plotsjl toolkit
  executable: julia
  command_line_arguments: --threads auto --optimize=0
  • Fixed an issue where invoking the plotsjl toolkit on Windows would sometimes fail with the error: Unknown system error 50.
  • Fixed an issue with R-based toolkits on Windows not being detected properly.


  • Added the ability to change the “Source code” label to other languages via configuration.
  • Added syntax highlighting to the linked source code.
  • Fixed an issue where code blocks with unicode symbols (e.g. greek letters) would trip up pandoc-plot (#16).


  • Added support for Pandoc 2.11. Unfortunately, there is no way to support prior versions of Pandoc at the same time.
  • With release, pandoc-plot has stabilized. The Haskell library API will not change until version 2+.


  • Fixed an issue where the current working directory was changed. This prevented users from referring to files in scripts with relative paths (#2).


  • Added executable caching: repeated usage of a particular toolkit will be faster because executables are only looked-for once.
  • Reverting the change from internal machinery of pandoc-plot has been moved to the Text.Pandoc.Filter.Plot.Internal module, where there is no guarantee of backwards-compatibility after 1.0.0.
  • Removed the makePlot function, which could not take advantage of multithreading and other key features of pandoc-plot.
  • Fixed an issue where files required for tests were missing from source tarballs (#13).


  • Fixed an issue where executables located on paths with spaces would not be invoked correctly (#12).
  • Fixed an issue where R-paths were not normalized correctly.
  • Fixed an issue where executables specified in configuration that did not exist would crash pandoc-plot.
  • Fixed an issue where some R-based toolkits appeared to be available, but were not.


  • Added the file parameter, which allows the user to read figure content from a file instead of using the code block content in documents. This is especially useful for complex figures, where you might want to have the help of your tooling in an IDE, for instance. Here’s an example:


  • Better error messages when specifying logger verbosity.
  • Cleaning output directories with pandoc-plot clean now follows configuration values for logging.
  • Fixed an issue where configuration in metadata did not get parsed properly.


  • The bokeh toolkit now supports exporting plots as SVGs (#8).
  • Interactive plots relying on javascript scripts will now defer loading the scripts (#9).
  • Added the dependencies argument, which tells pandoc-plot what files are important to a particular figure (#10). If a file listed in dependencies changes (for example, a data file), pandoc-plot will re-render the associated figure.
  • Better heuristic to determine what bokeh plot to save. This allows the user to export plots like the bokeh.layouts module.
  • Added support for the dpi parameter in graphviz and mathematica.
  • Added support for MATLAB’s new exportgraphics function introduced in MATLAB 2020a. Older versions fallback to using saveas.


  • The module Text.Pandoc.Filter.Plot.Internal is no longer exposed; instead, everything relevant is exposed by the Text.Pandoc.Filter.Plot module.
  • Fixed an issue where script errors would be logged as debug messages.
  • Interactive plots are now embedded directly in output (#7).


  • Added a new output format, HTML, to produce interactive plots. Not all renderers support it. You can try with Plotly/Python and Plotly/R as follows:

```{.plotly_python format=html}
import as px
df =
fig = px.scatter_ternary(df, a="Joly", b="Coderre", c="Bergeron")

  • Added a new toolkit, bokeh. This toolkit can take advantage of the new HTML interactive output.
  • Added a new toolkit, plotsjl.
  • Separated the detailed information from and into a proper This is now the information which will be shown with pandoc-plot --manual.
  • Exposed the pandoc-plot version via Text.Pandoc.Filter.Plot.pandocPlotVersion.


  • Fixed an issue where the pandoc version was not parsed properly, giving rise to errors when running pandoc-plot.
  • Fixed an issue where logging errors were not always displayed.


  • Removed dependency on open-browser package.
  • Starting with this version, pandoc 2.8 and 2.9 are no longer supported due to a breaking API change in pandoc 2.10.
  • Executables are now built with GHC 8.10.1.


  • Better multi-threaded logging. Only one thread (the logging thread) performs IO on the log file. This prevents hang-ups when working on large documents.


  • Added documentation on using pandoc-plot with LaTeX documents as well.
  • Added preliminary support for logging to pandoc-plot. You can turn on this feature in the configuration as follows:
    # Possible verbosity values: debug, error, warning, info, silent
    # debug level shows all messages
    # error level shows all but debug messages, etc.
    verbosity: info
    # OPTIONAL: log to file
    # Remove line below to log to stderr
    filepath: log.txt
  • Removed dependencies turtle, temporary, deepseq, and data-default-class, resulting in improved build times by ~10%, and makes the executable smaller by 15-20%!


  • Made the functions availableToolkits and unavailableToolkits public.
  • Minor documentation fixes.
  • Executables are now built with GHC 8.8.3.


New toolkits:

  • Added support for the Plotly/R plotting library.
  • Added support for Graphviz.

Other changes:

  • The determination of which figures to re-render or not has been improved. For example, changing the caption will not trigger a re-render of a figure anymore.

  • pandoc-plot will look for executables more thoroughly.

  • pandoc-plot toolkits will now show the exact executable that is being used, if possible.

  • Added a check when running the filter that the Pandoc version is at least 2.8. This is easier to understand that the default Pandoc warning on API incompatibility.

  • Added the ability to write the example configuration to an arbitrary file using pandoc-plot write-example-config.

  • Added the possibility to specify the configuration file via metadata. For example, in Markdown:

    title: My document
    author: John Doe
    plot-configuration: /path/to/file.yml

or on the command line:

pandoc --filter pandoc-plot -M plot-configuration=/path/to/file.yml ...
  • Added the ability to specify configuration file to the pandoc-plot clean and pandoc-plot toolkits commands.


  • The pandoc-plot executable will now process documents in parallel. This should dramatically speed up processing of large documents with lots of figures. This happens automatically through the function plotTransform.
  • Added a benchmarking suite.
  • Added defaultConfiguration so that people don’t have to install the data-default package to get access to default configuration values.
  • Added a check for the matplotlib toolkit, preventing users from using in figures. This would halt pandoc-plot.


  • Fixed an issue where the pandoc-plot executable could not be built outside of its git repository.


  • Updated documentation.
  • Added a --full-version flag to the executable, which includes which version of pandoc/pandoc-types was used, as well as the git revision.
  • Added the clean command to the executable. This can be used to clean-up output files produced by pandoc-plot.
  • Changed the flag --write-example-config to the command write-example-config.
  • Added the top-level function cleanOutputDir to clean output of pandoc-plot. This is only accessible if pandoc-plot is used as a library.
  • Added a distinction between failure to render a figure because of a mistake, and failing to render a figure because the toolkit is not installed. pandoc-plot will give better error messages in the latter case.


  • Added more examples.
  • Added MacOS binaries built via Azure pipelines.
  • BREAKING CHANGE: Parsing captions based on source file was not working. Captions format can be specified in the configuration file. This unfortunately changes the type signature of a few high-level functions.


  • Fixed an issue where paths with spaces would not work (issue #2).
  • Added Linux binaries built via Azure pipelines.


  • Improved documentation.


  • Added support for gnuplot.

  • Added more tests for all toolkits.

  • Fixed an issue where the package could not be installed because a source file was not included in the cabal file.


  • Initial release