Giving auto-completion candidates for dynamically typed languages requires complex analysis of the source code, especially when the goal is to ensure that the completion candidates do not introduce bugs. In this paper, we introduce an approach that builds upon abstract interpretation and the scope graph framework to obtain an over-approximation of the name binding seen at run-time. The over-approximation contains enough information to implement auto-completion services such that the given suggestions do not introduce name binding errors. To demonstrate our approach, we compare the suggestions given by our approach with the state of the art completion services on a subset of the Python programming language.