Is their any way to get custom visualizations into Metabase?

I'd like to expand on the visualization capabilities of Metabase by using external tools, such as Python.
At first I thought something like a "Jupyter Notebook Card" would be suffice, where it has a default variable named "data" that just references any table or model. Someone could write Python in the Notebook and have the output displayed, and optionally hide the code portion from view. Alas, I don't want to ask for a new feature here.

I'm more curious about now I could theoretically do this right now.
I know that Seaborn (the data visualization library) will create a temp file on your PC anytime you make a plot. When displaying the plot, it references that temp file. So, I think the easiest way to do this would be to use the Markdown Textbox (within Dashboards) to display that temp file.

Alas. then the question becomes, how do I get Metabase to trigger the creation of that temp file?

From that thread, the solution seems to be:

  1. Set up a server to run python and perform all the data wrangling / plot making
  2. Store the result in a database
  3. Connect Metabase to that database
  4. Source that database inside dashboards.

So, adapting this to my use case, I think I would need to:

  1. Set up a server to run python and perform all the data wrangling / plot making
  2. Store the result in a somewhere -- doesn't have to connect to Metabase
  3. Expose an endpoint to get the results/plot via URL (where Metabase will have access)
  4. Reference the image url in a Metabase Textbox

I can probably use a url query parameter to force the server to re-make the image--providing that "trigger" effect.

I think that could work. Though, I would really appreciate the input of someone more seasoned than myself. Would a different approach be ideal?

Thanks.