New repository: a set of JupyterLab commands for use with AI agents

[jtpio]

https://github.com/jtpio/jupyterlab-ai-commands defines a couple of extra "agents friendly" JupyterLab commands for working with the files and notebooks, since they work on the sharedModel directly.

For now the current commands are limited to file and notebook operations, but the idea would be to add more commands to that repo to enable more functionalities to agents.

Currently jupyterlab-ai-commands is useful for fully browser-based AI workflows, such as https://github.com/jupyterlite/ai.

But the current state of jupyterlab-ai-commands can also be combined with https://github.com/jupyter-ai-contrib/jupyter-server-mcp and https://github.com/jupyter-ai-contrib/jupyterlab-commands-toolkit, to make agents like Claude Code are able to discover such commands and execute them.

So it feels like this repo would belong to this org, as the JupyterLab / frontend equivalent of https://github.com/jupyter-ai-contrib/jupyter-ai-tools.

[jtpio]

@jupyter-ai-contrib/decision-makers

[dlqqq]

Would it be possible to eventually unify our tools & MCP servers under a single repo? I'm already getting lost between jupyterlab_commands_toolkit, jupyter_ai_tools, jupyter_server_ai_tools.

We can build these repos separately for now & plan to unify them later in a separate proposal. Just want to make sure we're all thinking of a unified outcome in the long-term.

[jtpio]

Not sure about unifying them, unless we make the repo a monorepo and have these packages released separately.

For the jupyterlite-ai use case, I'm only interested in the JupyterLab commands, not the server tools. So only the jupyterlab-ai-commands package.

[jtpio]

I've also been using jupyterlab-commands-toolkit / jupyter-server-mcp together with jupyterlab-ai-commands and Claude Code, so it would be great to keep that flexibility.

[dlqqq]

I'm OK with this proposal as long as we agree to unify our tool implementations in the long term.

NVM, Jeremy & Brian's replies make sense.

cc @jtpio @ellisonbg

[ellisonbg]

Given the goals of jupyter-ai-contrib to create modular building blocks, I think some level of "separate repo/package" makes sense. However, I do think we should aim to have one package for each such building block. So, for right now, both both have tools in jupyternaut and in some of the above repos - we should definitely unify those.

Question: what is the difference between jupyter-ai-tools and jupyter-server-ai-tools?

Another question/idea: do we want to name our repos that have actual tool implementations using the toolkit suffix to help users understand which repos have toolkits?

[Zsailer]

jupyter-server-ai-tools should go away. I think this is an old repo that just exposed tools through a server extension.

[ellisonbg]

Can we just archive it now?