That's nicely done but it is failing tests. We may want to mock the API call in the test. We generally don't want the tests to rely on an internet connection.

I got a chance to play around with the functions a bit more. I also meant to say "space group" instead of "phase" earlier and in the code. I think the issues are:

1. pymatgen removed some densities by default according to this: Deprecate CifParser.get_structures() in favor of new parse_structures in which primitive defaults to False materialsproject/pymatgen#3419, so it's missing a few common ones like SiO2. I'm wondering if it is better to call MP API directly, just so we have some fallback/default values?
2. It looks like there can be multiple densities for the same sample composition and space group. Not sure how to pick a default in that case? I think the order is a bit random and that's why the test failed here. We could also give users the option to specify space group & ID?

honestly I am not sure I would rely on pymatgen/MP. Why not just make an API call to COD and get the cif and compute the density?