This paper introduces MPCC++, a Model Predictive Contouring Control (MPCC) method designed for time-optimal flight with safety constraints in drone racing. The key contributions of MPCC++ include: 1. **Safety Constraints**: MPCC++ introduces a track constraint and a terminal set to prevent gate collisions and ensure recursive feasibility. The track constraint is a prismatic tunnel that joins the inner corners of the gates, while the terminal set is a periodic feasible trajectory that passes through the center of the gates. 2. **Dynamics Augmentation**: The nominal dynamics of the controller are augmented with a residual term to capture unmodeled effects such as aerodynamic forces and torques. This term is inferred from real-world data using polynomial regression. 3. **TuRBO Tuning**: The controller parameters are tuned using Trust-Region Bayesian Optimization (TuRBO), which efficiently explores the parameter space and balances exploration and exploitation. The paper compares MPCC++ against two baselines: MPCC and RL, and demonstrates its superior performance in both simulation and real-world experiments. MPCC++ achieves a 100% success rate in real-world experiments, outperforming MPCC and achieving similar lap times to the best-performing RL policy. The method provides an intuitive way to trade off performance and safety by adjusting the width of the tunnel. However, MPCC++ has limitations, including the need for a known centerline, longer training times compared to RL, and the assumption of known gate positions. Future work should address these limitations and explore incorporating gate position uncertainty.This paper introduces MPCC++, a Model Predictive Contouring Control (MPCC) method designed for time-optimal flight with safety constraints in drone racing. The key contributions of MPCC++ include: 1. **Safety Constraints**: MPCC++ introduces a track constraint and a terminal set to prevent gate collisions and ensure recursive feasibility. The track constraint is a prismatic tunnel that joins the inner corners of the gates, while the terminal set is a periodic feasible trajectory that passes through the center of the gates. 2. **Dynamics Augmentation**: The nominal dynamics of the controller are augmented with a residual term to capture unmodeled effects such as aerodynamic forces and torques. This term is inferred from real-world data using polynomial regression. 3. **TuRBO Tuning**: The controller parameters are tuned using Trust-Region Bayesian Optimization (TuRBO), which efficiently explores the parameter space and balances exploration and exploitation. The paper compares MPCC++ against two baselines: MPCC and RL, and demonstrates its superior performance in both simulation and real-world experiments. MPCC++ achieves a 100% success rate in real-world experiments, outperforming MPCC and achieving similar lap times to the best-performing RL policy. The method provides an intuitive way to trade off performance and safety by adjusting the width of the tunnel. However, MPCC++ has limitations, including the need for a known centerline, longer training times compared to RL, and the assumption of known gate positions. Future work should address these limitations and explore incorporating gate position uncertainty.