A　semi-active　strategy　for　model　predictive　control　(MPC),　in　which　magneto-rheological　dampers　are　used　as　an　actuator,　is　presented　for　use　in　reducing　the　nonlinear　seismic　response　of　high-rise　buildings.　A　multi-step　predictive　model　is　developed　to　estimate　the　seismic　performance　of　high-rise　buildings,　taking　into　account　of　the　effects　of　nonlinearity,　time-variability,　model　mismatching,　and　disturbances　and　uncertainty　of　controlled　system　parameters　by　the　predicted　error　feedback　in　the　multi-step　predictive　model.　Based　on　the　predictive　model,　a　Kalman-Bucy　observer　suitable　for　semi-active　strategy　is　proposed　to　estimate　the　state　vector　from　the　acceleration　and　semi-active　control　force　feedback.　The　main　advantage　of　the　proposed　strategy　is　its　inherent　stability,　simplicity,　on-line　real-time　operation,　and　the　ability　to　handle　nonlinearity,　uncertainty,　and　time-variability　properties　of　structures.　Numerical　simulation　of　the　nonlinear　seismic　responses　of　a　controlled　20-story　benchmark　building　is　carried　out,　and　the　simulation　results　are　compared　to　those　of　other　control　systems.　The　results　show　that　the　developed　semi-active　strategy　can　efficiently　reduce　the　nonlinear　seismic　response　of　high-rise　buildings.