The industry that is experiencing significant development is the music industry. An example of its development is the many online music service providers of application platforms. The amount of data stored makes it difficult to analyze existing data, the presence of Machine Learning is felt to be able to answer these challenges. Improving user experience is important to attract users to use the applications they have. The recommendation system is one way to improve that. This research aims to create a system that can present music recommendations according to user preferences so that the user's comfort level will increase. The system developed in this research uses the Extreme Programming method with several stages, namely planning, design, coding, and testing. This research utilizes Machine Learning in searching for data patterns and Content-Based Filtering (CBF) methods in finding recommendations. The recommendation system with the CBF method can produce a song similarity level of up to 0.6684, as well as the value of precision reaching 0.125 and 0.200 at recall. The results of Performance Testing and System Testing obtained stated that the recommendation system can run well with an average response time 3.5 seconds. The conclusion of this research is that the recommendation system using the CBF method can produce recommendations that are in accordance with user preferences, but with not too much data. More effective algorithms are needed for larger data.