We explore the potential for humanoid robots to interact with humans in hand-clapping games. In this context, a robot is able to adapt to random hand motions of humans in a timely fashion without preplaned information. This capability is built through: (1) predict human future motions in real time at very early stages (2) prediction involved in the robot dynamic systems for continuous movement control. We use Probability Movement Primitives (ProMPs) for human motion prediction and improved the accuracy through a motion recognition process with Heininger distance. To encode the possibility region of future human motions, a implicit Dynamic Movement Primitives (DMPs) is generated capturing different dynamics on one short for robot motion model. At last Model Predictive Controller (MPC) is applied to track K-step forward human motions to achieve time synchronization and joint goals. We present the results obtained for various hand-to-hand contacts between NAO robot and kids.