Dynamical mean-field theory has recently attracted much interests to provide a unified framework for the description of many aspects of nuclear dynamics [1, 2, 3, 4, 5] (for recent reviews see [6, 7]). In particular, the inclusion of pairing correlation has opened new perspectives [8, 9, 10, 11, 12]. A summary of recent applications including giant resonances and transfer reactions will be made in this talk [13, 14, 15, 16]. While new progresses have been made with the use of sophisticated effective interactions and the development of symmetry unrestricted applications, mean-field dynamics suffer from the poor treatment of quantum fluctuations in collective space. As a consequence, these theories are successful in describing average properties of many different experimental observations but generally fail to account realistically for the width of experimental distribution. The increase of predictive power of dynamical mean-field theory is facing the difficulty of going beyond the independent particle or quasi-particle picture. Nevertheless, in the last decade, novel methods have been proposed to prepare the next generation of microscopic mean-field codes able to account for both average properties and fluctuations around the average. A review of recent progresses in this direction as well as recent applications to heavy-ion collisions will be given [17, 18].