This paper presents MPCC++, a model predictive contouring control (MPCC) method for time-optimal drone racing with safety constraints. The method introduces three key components to enhance the state-of-the-art MPCC approach: safety guarantees through track constraints and terminal sets, augmented dynamics with residual terms capturing aerodynamic effects, and Trust-Region Bayesian Optimization (TuRBO) for hyperparameter tuning. The proposed approach achieves similar lap times to the best-performing reinforcement learning (RL) policy and outperforms the best model-based controller while satisfying constraints. In both simulation and real-world experiments, the method consistently prevents gate collisions with 100% success rate, pushing the quadrotor to its physical limits with speeds exceeding 80 km/h. The paper introduces a track constraint as a prismatic tunnel that prevents gate collisions while allowing time optimization. The terminal set consists of a periodic, feasible trajectory, ensuring recursive feasibility and inherent robustness. The method augments the nominal dynamics with a residual term that captures unmodeled effects, such as aerodynamic forces, inferred from real-world data. TuRBO is used to tune the MPCC controller's hyperparameters based on lap time minimization, resulting in superior performance compared to previous work. The method is tested on the Split-S race track with 7 gates across three simulation environments: a simple simulator, a high-fidelity simulator using Blade Element Momentum Theory (BEM), and a data-driven simulator predicting aerodynamic forces from real-world data. The method is also validated in the real world using a high-performance racing drone. In real-world experiments, the method achieves 100% success rate with speeds exceeding 80 km/h, outperforming the baseline MPCC and achieving similar lap times to the best-performing RL policy. The method's safety guarantees and ability to handle model mismatches and disturbances make it a robust solution for drone racing.This paper presents MPCC++, a model predictive contouring control (MPCC) method for time-optimal drone racing with safety constraints. The method introduces three key components to enhance the state-of-the-art MPCC approach: safety guarantees through track constraints and terminal sets, augmented dynamics with residual terms capturing aerodynamic effects, and Trust-Region Bayesian Optimization (TuRBO) for hyperparameter tuning. The proposed approach achieves similar lap times to the best-performing reinforcement learning (RL) policy and outperforms the best model-based controller while satisfying constraints. In both simulation and real-world experiments, the method consistently prevents gate collisions with 100% success rate, pushing the quadrotor to its physical limits with speeds exceeding 80 km/h. The paper introduces a track constraint as a prismatic tunnel that prevents gate collisions while allowing time optimization. The terminal set consists of a periodic, feasible trajectory, ensuring recursive feasibility and inherent robustness. The method augments the nominal dynamics with a residual term that captures unmodeled effects, such as aerodynamic forces, inferred from real-world data. TuRBO is used to tune the MPCC controller's hyperparameters based on lap time minimization, resulting in superior performance compared to previous work. The method is tested on the Split-S race track with 7 gates across three simulation environments: a simple simulator, a high-fidelity simulator using Blade Element Momentum Theory (BEM), and a data-driven simulator predicting aerodynamic forces from real-world data. The method is also validated in the real world using a high-performance racing drone. In real-world experiments, the method achieves 100% success rate with speeds exceeding 80 km/h, outperforming the baseline MPCC and achieving similar lap times to the best-performing RL policy. The method's safety guarantees and ability to handle model mismatches and disturbances make it a robust solution for drone racing.