Recent evolution of technology transformed the way we interact with Information Systems (IS), leading to a new generation of IS, the Pervasive Information Systems (PIS). These systems have to face heterogeneous pervasive environments, whose complexity they must hide from end-user. In order to reach transparency and proactivity necessary for successful PIS, new discovery and prediction mechanisms are necessary. In this paper, we propose a new user-centric approach for service discovery and prediction on PIS based on both user's context and intentions. Intentions allow focusing on goals user wants to satisfy when requesting a service. Those intentions rise in a given context, which may condition the service implementation. We propose then a service discovery mechanism that observes user's context and intention in order to offer him the service that may best satisfy her/his intention on the current context. We also propose a prediction mechanism that tries to anticipate user's intentions considering the observed context and user's history.