Recommending a certain number of services based on QoS (Quality of Service) is a very widespread problem. Traditional approaches use the QoS advertised by service providers to calculate the top-k services or the skyline ones. The drawback of the top-k operator is that it requires users to provide a utility function, which is a rather demanding task. In turn, the skyline operator provides a large number of services, which makes difficult taking a quick/good decision from the user's side. Recent approaches rely on user-service interactions to predict QoS of the services, which suffer from the well-known cold start problem. In this paper, we propose a clustering-based diversity approach for service recommendation. Similar to the top-k operator, our approach assumes that the users have some utility functions, but it does not ask them to provide their utility functions. Like the skyline operator, it selects only skyline services; however, it does not overwhelm users by recommending a large number of services. We demonstrate the both effectiveness and efficiency of our approach on two real datasets.