Recommendation technologies are being increasingly adopted by mobile applications. Multiple algorithms and strategies are proposed to recommend the best points of interest to users of handheld devices. Unfortunately, developers of these systems receive little support to deal with the challenging integration of such approaches into a software development process. In this work, we introduce a conceptual framework to develop mobile recommender systems of geographic points of interest, from a model-driven perspective. This framework is composed of three phases that help developers specify the structural, behavioral and navigational aspects of a recommender system from a high abstraction level. In this way, our proposal supports the implementation of collaborative filtering algorithms and their combination with content-based and location-based constraints, based on multiple preference actions usually performed in the evaluation of geographic locations. A case study is introduced to illustrate this proposal.