This paper addresses a constrained model predictive control (MFC) scheme based on the disturbance observer (DOB) that can handle constraints on the total control input. In a DOB-based MFC scheme, total control input is the sum of the DOB output and MFC output, however, it is not possible to impose constraints on the total input in a conventional DOB-based MFC scheme. We therefore bring a novel concept called time-variant input constraints into our design and construct a useful structure of combined MFC-DOB systems. The proposed DOB-based MFC scheme is applied to a position control problem of a two-link manipulator, and its effectiveness is checked through an experimental evaluation.