Service compositions, e.g., Internet of Things (IoT) compositions or RESTful service compositions are widely used in the industry to enhance the interoperability and integration of their systems and applications. Testing service compositions is long and difficult activity as each service may be deployed on different servers and often requires specialised testing tools. In order to help developers in this activity, this paper proposes an automated approach to generate test cases for experimenting every service in isolation. These test cases can be later adapted or used for regression testing. This approach is based upon 4 steps that aim to: 1. extract traces from event logs, 2. gather similar behaviours to reduce the final test case number and to extract knowledge that will be used during the test case generation, 3. produce generic test cases given under the form of IOTS (Input Output Transition Systems) that encode the use of mock components and provide test verdicts, 4. generate test scripts and mock components for every testable service. We report experimental results obtained from 4 case studies and show that our algorithms build effective test cases and scale well with the event log size.