In this paper, an inexact nonlinear model predictive control scheme with reduced computational complexity is proposed. The presented approach exploits fixed sensitivity information precomputed offline at a reference value. This allows one to avoid the online computational effort resulting from the propagation of sensitivities and possibly the corresponding condensing routine when solving the optimal control problem with a sequential quadratic programming method. By performing a numerical simulation of the nonlinear dynamics online, feasibility of the closed-loop trajectories can be preserved in contrast to linear model predictive control schemes. Nominal stability guarantees of the approach are derived and the effectiveness of the scheme is demonstrated on a non-trivial example.