Humanoid robots are commonly required to make contacts with the environment in order to perform various tasks. For tasks such as walking, sudden establishment and breaking of contacts are unavoidable. Such behaviours may result in large changes in actuator torques (lashing), leading to potential control instability and damage to the actuators. In this paper, an approach to minimise sudden changes in actuation torques is proposed. This is achieved using Model Predictive Control (MPC) to modify the maximum and minimum allowable forces for a reactive whole-body controller. The MPC previews the evolution of contacts in time and generates smoother maximum and minimum allowable forces. The effect of the proposed approach is simulated for different scenarios on the iCub robot, such as standing up from a sitting posture and the lifting and lowering of a foot while standing. The results show that the proposed approach significantly decreases the changes in joint torques at the instances when contacts are broken or established.