Mobile-edge computing (MEC) could support edge caching and processing for video service, which essentially changes “cache and transmit” mechanism and provides another video retrieval mode. However, existing caching policies are usually inefficient, mainly due to lack of: 1) interaction of video caching and retrieving which happen in different timescales and 2) collaboration among servers. As motivated, we jointly consider long-term caching placement and short-term video retrieval mode selection in coordinated multi-server MEC systems to enrich user experience. Besides, we propose a sample average approximation-based two-phase algorithm which outmatches existing approaches in terms of content access delay and hit ratio.