A feedback predictive controller design method is proposed for a class of uncertain discrete-time Markov jump systems. As engineering application is considered, the polytope uncertainty exists in dynamic parameters under each mode of jump systems is introduced, while partly unknown transition probabilities between jump modes are also investigated. A predictive controller with corresponding invariant ellipsoid is obtained by minimizing infinite horizon quadratic performance index. The controller guarantees the closed-loop system robust mean square stable. By applying the obtained feedback
controller gain, a set of asymptotically stable invariant polyhedral can be constructed on line, which enlarges the admissible set of the state in some extent. The numerical example verifies the feasibility and effectiveness of the proposed method.