In this paper a collaborative music game for two pen tablets is studied in order to see how two people with no professional music background negotiated musical improvisation. In an initial study of what it is that constitutes play fluency in improvisation, a music game has been designed and evaluated through video analysis: A qualitative view of mutual action describes the social context of music improvisation: how two people with speech, laughter, gestures, postures and pauses negotiate individual and joint action. The objective behind the design of the game application was to support players in some aspects of their mutual play. Results show that even though players activated additional sound feedback as a result of their mutual play, players also engaged in forms of mutual play that the game engine did not account for. These ways of mutual play are descibed further along with some suggestions for how to direct future designs of collaborative music improvisation games towards ways of mutual play.