Summary: This paper considers the constrained control problem of discrete-time stochastic systems with multiplicative uncertainty. We design a stochastic model predictive control algorithm based on the formulation of multi-layer probabilistic sets. Multi-layer probabilistic sets describe the distribution regions where the system evolves with different probabilities under multi-step feedback laws, thus enabling the satisfaction of soft constraints at different probabilistic levels. This algorithm has a large applicable region by dynamically optimizing multi-step feedback laws. Furthermore, we propose a simplified algorithm that reduces the computational burden with a guarantee of its applicable region.