This paper tackles the challenging issue of collision avoidance among multiple dynamical agents, a problem known for its non-convex nature and computational complexity. Our proposed solution involves decomposing the space into time-varying Voronoi cells with buffer zones, exploiting the versatility of B-splines to reconfigure trajectories when collisions are predicted. This approach is integrated into a Model Predictive Control (MPC) framework, implemented in a distributed manner where agents communicate predicted control point sequences defining the B-spline curves. Through simulation testing across various scenarios, our approach demonstrates significant promise, paving the way for experimental validation and real-world applications.