We propose a KeyGraph-based context sharing method in mobile environment. With the recent advancement of mobile sensors, a variety of mobile applications become vehicles for improving our lives. Context sharing system which shares the user behaviors, emotion, and location is one of the promising fields for the social network service. It is a difficult problem to determine whether a user will share the personal information or not. In typical social network models, users are grouped in communities, and nodes of the same community have strong social links between each other. However, some nodes also have social links outside their "home" community. They have social relationships with users of different groups. Most systems concentrate on generating internal "home" community regardless of outside social relation. In this paper, we classify the personal information into two types. First type is the information to be shared with "home" community only. Second type is the information to be shared with as many people as possible. We utilize KeyGraph algorithm to select a home community for sharing the personal contexts. KeyGraph extracts two types of people who have strong social relationships in a community and have social links with many different communities. In order to show the feasibility of the proposed method, we conduct experiments to extract the user communities from Bluetooth data and implement a real-time context sharing application.