Input constraints present difficulty in the design of stable controllers. A convenient way to stability is to ensure short-term feasibility, that is to ensure the existence of a set of future controls which are within input constraints and meet suitable terminal constraints. Earlier algorithms use terminal constraints which depend on the setpoint and hence short-term feasibility can be lost due to excessive setpoint changes. In the paper a simple algorithm is proposed which retains short-term feasibility by allowing the terminal constraint to change, but still guarantees convergence while at the same time optimising a cost function which is related to actual input/output predicted performance.