||Model Predictive Control (MPC) is an optimal control approach where constrained optimization problems are solved iteratively over receding horizons on-line, effectively combining prediction of future system behavior and constraints handling, which are both relevant features for attaining high performance in legged locomotion. In this work, a Model Predictive Control (MPC) framework for walking which permits automatic footstep placement is extended to also allow footstep duration decision. In the proposed scheme, mixed-integer quadratic programs (MIQP) are solved on-line to simultaneously decide center of mass jerks, footsteps positions and steps durations while respecting actuation, geometry, and contact constraints. Simulations results using the linear inverted pendulum model show that deciding both footstep placement and timing expands the disturbance tolerance margin provided by a controller that selects footsteps positions but has fixed step duration. Moreover, results using a full-body simulation model of the Robotis OP2 robot are presented to validate the proposed controller in a realistic environment. To reduce the computational burden, formulations where the robot may decide only the duration of the first footstep are also investigated, showing computational time amenable to real-time implementation.