Although remarkable progress has been made by the existing federated learning (FL) platforms to provide fundamental functionalities for development, these FL platforms cannot well satisfy burgeoning demands from rapidly growing FL tasks in both academia and industry. To fill this gap, in this paper, we propose a novel and comprehensive federated learning platform, named FederatedScope, which is based on a message-oriented framework. Towards more handy and flexible support for various FL tasks, FederatedScope frames an FL course into several rounds of message passing among participants, and allows developers to customize new types of exchanged messages and the corresponding handlers for various FL applications. Compared to the procedural framework, the proposed message-oriented framework is more flexible to express heterogeneous message exchange and the rich behaviors of participants, and provides a unified view for both simulation and deployment. Besides, we also include several functional components in FederatedScope, such as personalization, auto-tuning, and privacy protection, to satisfy the requirements of frontier studies in FL. We conduct a series of experiments on the provided easy-to-use and comprehensive FL benchmarks to validate the correctness and efficiency of FederatedScope. We have released FederatedScope for users on https://github.com/alibaba/FederatedScope to promote research and industrial deployment of federated learning in a variety of real-world applications.