Model predictive control (MPC) is an effective way to reduce air-conditioning energy use and improve demand response within a predicted horizon while maintaining indoor thermal comfort. In office buildings with variable air volume (VAV) systems, personalized MPC can be achieved by varying the temperature constraints in different thermal zones based on the thermal preferences of occupants. This paper explores the dynamic temperature constraints set in personalized MPC, to strike a balance between personal thermal comfort and energy cost for air-conditioning. A co-simulation framework of EnergyPlus and an add-on MPC controller is constructed, using a prototype office building as a testbed. A Bayesian-based probabilistic model is built to represent each occupant’s thermal preference. Dynamic temperature constraints are set in the model predictive controller based on personal thermal profiles, variable utility rates (energy cost) and weather conditions. The personalized MPC with dynamic setpoint constraints can achieve an optimal balance between maintaining thermal comfort and reducing energy cost during the cooling season, with 20% comfort improvement vs. typical MPC with static setpoint constraints, while the extra energy cost is only $0.2/day.