We propose and study the Maximum Constrained Agreement Sub- tree (MCAST) problem, which is a variant of the classical Maximum Agreement Subtree (MAST) problem. Our problem allows users to ap- ply their domain knowledge to control the construction of the agreement subtrees in order to get better results. We show that the MCAST prob- lem can be reduced to the MAST problem in linear time and thus we have algorithms for MCAST with running times matching the fastest known algorithms for MAST.