A Model Predictive Control (MPC) strategy for nonlinear Networked Control Systems (NCS) described by a polynomial plants subject to input, state constraints and data-losses is introduced. The results here proposed represent a significant extension to the polynomial framework of a recent Receding Horizon Control (RHC) scheme developed for input-saturated multi-model systems. Sum-of-Squares (SOS) conditions for polynomial systems are derived together with an SDP formulation of the on-line optimization problems.