Skip to content
Jupyter+DAR compatibility exploration
Jupyter Notebook Python JavaScript HTML
Branch: master
Clone or download

Latest commit

Fetching latest commit…
Cannot retrieve the latest commit at this time.


Type Name Latest commit message Commit time
Failed to load latest commit information.

Jupyter + Stencila = nbstencilaproxy

Jupyter + Dar file format compatibility exploration for running Stencila editor on Binder

PyPI version


Click on the button below to launch an online Jupyter instance on based on this repository:


Open an the example Dar project by clicking on "New > Stencila Session":


This project is part of the eLife Innovation Sprint 2018 and Mozilla Global Sprint 2018 (see

The projects core module is a Python package, nbstencilaproxy, which is available on PyPI. It is based on nbserverproxy to run the services and the UI required to edit Stencila documents in the browser. The concept is inspired by nbrsessionproxy.

nbstencilaproxy includes the following components:

  • a JavaScript package for installing and running Stencila (client/UI and services) in a Jupyter container
  • extension for the Jupyter UI to add a "New Stencila Session" button


How does it work?

The building blocks

Binder relies on repo2docker create a runtime environment from source code repositories. repo2docker was extended as part of this project to detect Dar documents using the file manifest.xml, which is contained in all Dars. If a manifest.xml is found, then repo2docker installs nbstencilaproxy into the image (see code) and creates environment variables (STENCILA_ARCHIVE_DIR and STENCILA_ARCHIVE) to configure a single Dar document to be viewed in the Stencila UI.

The installation consists of the following steps and components:

  1. Install Stencila for Python from GitHub and register the package (thanks to the registration, the Stencila Host can find and connect these runtimes)
  2. Install nbstencilaproxy itself, including a JavaScript module (see below for description of the npm; see for installation of the npm package as part of the Python module)
  3. Install and enable the nbstencilaproxy Jupyter notebook server extension, which serves the Stencila UI and services via proxies (see nbstencilaproxy/
  4. Install and enable the nbstencilaproxy Jupyter notebook extension, which adds a menu item into the Jupyter UI to open Stencila (see nbstencilaproxy/static/tree.js)

JavaScript module within nbstencilaproxy

The PyPI module includes a small npm package with JavaScript code, which serves the following purposes:

  • pulls in Stencila' JavaScript depenencies, namely stencila and stencila-host (see package.json)
  • includes code to run the dar-server and a static file server for the app using the files from the the module stencilas directory dist (see the file stencila.js for details)
  • run a stencila-host on the port provided by nbserverproxy (see stencila-host.js)

This gives us control of the paths and let's us get rid of complex development features (e.g. substance-bundler).

The package includes stencila-node as a dependency. It provides the JupyterContext as well as a NodeContext (for executing Javascript) and a SqliteContext (for executing SQL).

We also made our own version of app.js, getting rid of the virtual file storage stuff (vfs), defaulting storage to fs (file system), because that is what is needed for Jupyter - we do not need to host any examples. In the same line, we built own index.html (based on example.html) and serve that, which allows us to directly render a Dar document instead of a listing of examples and instruction and to use app.js.

Relevant path configurations comprise (a) the local storage path and (b) the URLs used by the client, accessing the dar-server through nbserverproxy.

Connecting Stencila to Jupyter kernels

Stencila has "execution contexts" (the equivalent of Jupyter's "kernels") for R, Python, SQL, JavaScript (Node.js), and Mini (Stencila's own simple language). Execution contexts differ from kernels in a number of ways including local execution and dependency analysis of cells. Both of these are necessary for the reactive, functional execution model of Stencila Articles and Sheets.

It is possible to install these execution contexts in the Docker image, and is done so for R However, Stencila also has a JupyterContext which acts as a bridge between Stencila's API and Jupyter kernels. Since the base image for BinderHub already has a Jupyter kernel for Python installed, it can always be used. This does mean however, that some of the reactive aspects of the Stencila UI won't work as expected. Also the JupyterContext is not well developed or tested.

Install locally

If you have a local Jupyter instance, you can install the package directly:

pip install nbstencilaproxy

Enable the extensions for all users on the system:

jupyter serverextension enable  --py --sys-prefix nbstencilaproxy
jupyter nbextension     install --py --sys-prefix nbstencilaproxy
jupyter nbextension     enable  --py --sys-prefix nbstencilaproxy

Run locally with repo2docker

You can run the latest release of nbstencilaproxy as part of repo2docker.

# install repo2docker:

# run repo2docker for a repository with Dar directories, e.g. archive/ in this repository (add '--no-build' option to only inspect Dockerfile)
repo2docker --debug ./archive
  • Login by visiting the tokenized URL displayed e.g. http://localhost:8888/?token=99a7bc13...
  • Click on the "New > Stencila Session" button on the Jupyter start page, opening the py-jupyter example, or
  • Open one of the included examples by appending the following parameters to the URL:


Updating Stencila

The following places use/install Stencila components in specific versions:

  • repo2docker/buildpacks/ installs
  • repo2docker/buildpacks/ installs
  • nbstencilaproxy/package.json installs stencila-node and stencila from npm

Using local nbstencilaproxy in repo2docker

This is quite tricky to do, but useful to test the whole stack of tools. It could be possible to copy the local files into the container, but a more practical approach is to temporarily telling repo2docker to install nbstencilaproxy from ones own fork.

In repo2docker/buildpacks/ find the line with

${NB_PYTHON_PREFIX}/bin/pip install --no-cache nbstencilaproxy==0.1.1 && \

which installs nbstencilaproxy from PyPI. Replace it with the following line (using your username for <user>)

${NB_PYTHON_PREFIX}/bin/pip install<user>/nbstencilaproxy/archive/master.tar.gz && \


BSD 3-Clause License

You can’t perform that action at this time.