This paper shows a way to build a music recommendation system based on datamining algorithms and a neighbor graph. A extensible python library is developed, using a variety of inputs differing from music metadata like lyrics or the genre to the analysis of the associated audio data. In order to demonstrate and verify the results, a Gtk+–based MPD client is developed that can be used either as debugging tool or normal media player. Various techniques are shown to compare all possible attributes of a song with each other in an efficient way.