Direct model reference adaptive control (DMRAC) is an attractive algorithm that uses a linear combination of feedforward model states and command inputs, as well as the error between plant and model outputs; it does not require full state access nor observers. Real world applications are usually subject to limits on the controlled variables for either safety or practical reasons. Because of the unique structure of the controller, many of the methods developed to deal with saturation constraints can not be applied directly to DMRAC. This papers presents extensions to the algorithm to allow for the explicit handling of such constraints.