In situations where the structural parameters of a redundant manipulator are unavailable, the motion control strategies discussed in earlier chapters become inapplicable, making model-independent approaches particularly appealing. Given the benefits of model predictive control (MPC), including its inherent capability to directly manage constraints, this chapter constructs a model-free MPC algorithm tailored for redundant manipulators lacking structural information. This chapter utilizes a cerebellum-inspired framework constructed with an echo state network (ESN) to replace the kinematic model of the redundant manipulator and introduces an MPC algorithm that integrates this cerebellar model with a neural dynamics (ND) method. In contrast to prior research, this algorithm processes both the optimization of system performance and the incorporation of constraints for the redundant manipulator, achieving high-accuracy prediction and tracking through the development of an online learning algorithm for the cerebellar model. Additionally, this chapter introduces a correction mechanism grounded in ND to refine the prediction model, along with a discrete dynamic neural network (DDNN) model designed to implement the MPC framework. Theoretical evaluations validate the convergence properties of both the ND-based correction method and the DDNN model. Simulations and experiments consistently verify the effectiveness of the cerebellum-inspired MPC (CIMPC) algorithm, which surpasses existing methods in the tracking accuracy.