Previous works on cache-enabled small-cell networks (SCNs) with probabilistic caching often assume that each user is connected to the nearest small base station (SBS) among all that have cached its desired content. The user may, however, suffer strong interference from other SBSs which do not cache the desired content but are geographically closer. In this work, we investigate this issue by deploying multiple antennas at each SBS. We first propose a user-centric SBS clustering model where each user chooses its serving SBS only from a cluster of $K$ nearest SBSs with $K$ being a fixed cluster size. Two beamforming schemes are considered. One is coordinated beamforming, where each SBS uses zero-forcing (ZF) beamformer to null out the interference within the coordination cluster. The other is uncoordinated beamforming, where each SBS simply applies matched-filter (MF) beamformer. Using tools from stochastic geometry, we obtain tractable expressions for the successful transmission probability (STP) of a typical user for both cases in the high signal-to-noise ratio (SNR) region. Tight approximations in closed-form expressions are also obtained. We then formulate and solve the optimal probabilistic caching problem to maximize the STP. Numerical results reveal interesting insights on the choices of ZF and MF beamforming in multi-antenna cache-enabled SCNs.