jupyter-marimo-proxy enables the JupyterLab launcher and the classic Jupyter Notebook file browser to launch Marimo.

On a JupyterHub deployment, jupyter-marimo-proxy leverages JupyterHub's existing authenticator and spawner to launch Marimo within users' Jupyter environments.

jupyter-marimo-proxy requires Marimo, but does not explicitly declare a dependency on marimo, so they may be installed separately. Both may be installed using pip like so:

$ pip install 'marimo>=0.6.21' jupyter-marimo-proxy

The following Dockerfile builds an image that runs JupyterHub (on port 8000) with DummyAuthenticator (demo:demo), LocalProcessSpawner, Marimo, and jupyter-marimo-proxy.

FROM	quay.io/jupyterhub/jupyterhub:latest
RUN	cd /srv/jupyterhub && jupyterhub --generate-config && \
	echo "c.JupyterHub.authenticator_class = 'dummy'" >> jupyterhub_config.py && \
	echo "c.DummyAuthenticator.password = 'demo'" >> jupyterhub_config.py && \
	pip install --no-cache-dir notebook 'marimo>=0.6.21' jupyter-marimo-proxy
RUN	useradd -ms /bin/bash demo

With more complicated setups that include multiple Python environments, it is vital to determine where each package is to be installed. Marimo should be installed into the user's environment to access the user's packages but made available in the search path so Jupyter could find it, and jupyter-marimo-proxy must be installed directly into Jupyter's environment so Jupyter could import it.

Consider the following example, in which Jupyter comes pre-installed in the root environment but Miniforge is installed for the user. We take care to install Marimo using /opt/conda/bin/pip and jupyter-marimo-proxy using /usr/bin/pip. By the magic of search path manipulation, Marimo is also available to Jupyter.

FROM	quay.io/jupyterhub/jupyterhub:latest

RUN	cd /srv/jupyterhub && jupyterhub --generate-config && \
	echo "c.JupyterHub.authenticator_class = 'dummy'" >> jupyterhub_config.py && \
	echo "c.DummyAuthenticator.password = 'demo'" >> jupyterhub_config.py && \
	pip install --no-cache-dir notebook

ENV	PATH=/opt/conda/bin:$PATH
RUN	curl -fsSL https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh -o /root/miniforge.sh && chmod +x /root/miniforge.sh && \
	bash /root/miniforge.sh -b -p /opt/conda && rm /root/miniforge.sh

RUN	/opt/conda/bin/pip install --no-cache-dir 'marimo>=0.6.21'
RUN	/usr/bin/pip install --no-cache-dir jupyter-marimo-proxy

RUN	useradd -ms /bin/bash demo

jupyter-marimo-proxy provides search path modification by setting a single path for environment variable JUPYTERMARIMOPROXY_PATH.

Setting JUPYTERMARIMOPROXY_PATH to $HOME/.local/bin or $HOME/bin is not required. $HOME/.local/bin and $HOME/bin are automatically prepended at runtime if these directories exist.

JUPYTERMARIMOPROXY_PATH may need to be added to c.Spawner.env_keep in the JupyterHub configuration.

Marimo and jupyter-marimo-proxy should be installed into the single-user containers. They are not needed by the main hub.

Make sure jupyter-marimo-proxy is installed into the same Python environment where Jupyter is installed. See advanced example above.

Make sure Marimo is installed and available in the search path. If the search path were modified in a descendent of Jupyter, the modification would not be available to Jupyter itself. See advice regarding search path modification above.

b-data customers should use b-data's fork.

Make sure Marimo is installed into the Python environment where these modules are installed. Alternatively, make sure the expected modules are installed into the Python environment where Marimo is installed. See advanced example above.

The --base-url argument was introduced to marimo edit in version 0.6.21. Try Marimo 0.6.21 or newer.