SPURS is a protein stability predictor built by rewiring two pretrained protein generative models: a frozen ProteinMPNN structure encoder is wired into an ESM2-650M sequence decoder through a structural adapter, and a small MLP head reads the fused representation to emit a normalized folding free-energy change (ddG) for every possible amino-acid substitution at every position of a chain. A companion checkpoint (SPURS-multi) extends the same backbone with an amino-acid embedding and a second MLP head to score combinations of point mutations. The predicted ddG landscape, combined with ESM-1v wild-type-vs-mutant logit differences and a sigmoid regression, is also used to score how functionally important each residue position is.