Agents are often lead to make collective decision (reaching a consensus about task or resource allocation, election, etc). This paper proposes a distributed protocol based on iterative exchange of opinions among strongly autonomous, weakly rational and heterogeneous agents. We prove that the protocol converges to a consensus while respecting agents’ autonomy and fairness. First, a formalism to model agents’ preferences, positions and opinions is developed. Then several operators required by the protocol are characterized (e.g. opinion cycle detector, aggregation operator, and consensus detector). Finally, the main experimental results we obtained when the protocol has been applied to the coalition formation problem in the e-commerce context.