It is challenging to model and control a tail-sitter unmanned aerial vehicle (UAV) because its blended wing body generates complicated nonlinear aerodynamic effects, such as wing lift, fuselage drag, and propeller-wing interactions. We therefore devised a hybrid aerodynamic modeling method and model predictive control (MPC) design for a quadrotor tail-sitter UAV. The hybrid model consists of the Newton-Euler equation, which describes quadrotor dynamics, and a feedforward neural network, which learns residual aerodynamic effects. This hybrid model exhibits high predictive accuracy at a low computational cost and was used to implement hybrid MPC, which optimizes the throttle, pitch angle, and roll angle for position tracking. The controller performance was validated in real-world experiments, which obtained a 57% tracking error reduction compared with conventional nonlinear MPC. External wind disturbance was also introduced and the experimental results confirmed the robustness of the controller to these conditions.