Jupyter has been an extensible computing environment for many years. In JupyterLab, almost everything you see is a plugin, the whole UI is driven by commands, and kernels run code and return results. Jupyter Server can also be extended, adding new capabilities on the backend.

Coding agents can now work against those same pieces, which makes customizing and owning your computing environment more accessible than before.

First, agents make it easier to build your own Jupyter. What used to be a week of extension work or config adjustment can be an afternoon.

Second, the command registry is a control surface. Every action in the UI is already a command, so you can hand those commands to an agent, for example over MCP, and let it open files, run actions, and reshape the layout. The agent operates the environment rather than just typing into it.

Third, kernels give agents a place to run code and see the results, right next to a UI they can reshape, and can quickly produce live, interactive visualizations.

The result is an environment an agent can genuinely reshape and operate. To keep it concrete, we will demo existing examples of xtralab (a custom JupyterLab metapackage) and other workflows running right in the browser with JupyterLite.