The article introduces a circulating current reference generator based on a constrained model predictive control (MPC) in modular multilevel converters (M2C) for drive applications. It aims to control the energy in each cluster by generating optimal circulating currents over a long-horizon prediction while considering branch current limits. The optimal control problem is decomposed into smaller subproblems, which allows the utilization of an iterative algorithm based on the alternating direction method of multipliers (ADMM) to efficiently solve the optimization problem. As a result, the ADMM-based reference generator enables a considerably extended prediction horizon, which enhances M2C performance, particularly in balancing and mitigating low-frequency oscillations in capacitor voltages during challenging operating conditions like low-speed and high-torque scenarios. The proposed control is validated by experimental results using an M2C with four cells per cluster driving a 3 kW induction machine.