MCMC in Python: observed data for a sum of random variables in PyMC

I like answering PyMC questions on Stack Overflow, but sometimes I give an answer and end up the one with the question. Like what would you model as the sum of a Poisson and a Negative Binomial?