If using a standard formulation you can reduce the weight on the control action in the cost function. If formulating MPC from scratch, you can introduce a decay rate in the Lyapunov derivative to achieve exponential stability on the tracking error (with a tunable decay rate) as opposed to asymptotic stability.
It is totally dependent on your particular problem and formulation. For example, if your state equations have been derived in a linear form, you can use the various features of MPC toolbox in MATLAB. But, if you tend to produce a unique code for a nonlinear aggressive problem, it is needed to tune the penalty weighing matrices by some algorithm approaches like GA. In Explicit NMPC you should calculate the controlling inputs within an offline approach by using controlling references and quadratic cost-functions (mp-Lp) with an appropriate optimizer (fmincon). In this case this is very important to set your equations time step properly with regards to your system dynamics; then you can use the offline results in an online simulation with a low level controller (PID) and calculate the time response...