Interaction can take place in multi-agent systems (MAS), only if each agent contains features devoted to it. These features are usually difficult to implement and programmers have often to redesign them, even when they take inspiration from generic interaction protocols. Most often, interaction features are not separated from other, functional, ones and this places a limitation on the reusability potential of the agent’s interaction capabilities in different contexts. In such a situation inconsistencies are difficult to anticipate and solve. In this paper, we propose a method to address these issues by making separate but still consistent models for the coordination and the functional aspects of the agents and by deriving automatically the agent’s effective interaction model from generic interaction protocols. An implementation of this method is also described.