PyMC can definitely compute the logp for a single variable, when you call model.logp(vars=x). You can check if Slice can work like that or of it needs the whole model logp to make correct transitions.
I am not familiar with theory/ literature on using Slice for subsets of variables.