This paper studies the consensus for a group of agents which have a discrete-time double-integrator dynamics. To solve this problem, a Model Predictive Control (MPC) scheme is introduced. The proposed control protocol is decentralized and is designed by combining graph theory with a predictive control algorithm to take into account the switches on the communication topology. The predictive strategy is used to estimate input and output of agent through a receding horizon. Contrary to many existing works, the cost function is designed using the difference between two consecutive inputs. The controller has integrator properties to eliminate steady-state errors. Finally, some simulation results are given to show the effectiveness of the proposed model predictive control.