Multi-agent interaction protocols are used to support extended conversations between interacting agents. Traditionally, agent interaction protocols have just specified the overt interactions that take place among communicating agents and have ignored other, non-interactive, events that may have a bearing on the ensuing conversation. In this paper we present our own approach for modelling interaction protocols, which has drawn inspiration from narrative intelligence studies and which can incorporate aspects of the conversational context which are germane to the agent interaction. We argue that our approach is more suited to the support of complex interactions, such as can occur in e-business operations among agents, and we illustrate our approach by discussing an example based on commodities trading that involves multiple exchanges among a collection of interacting agents.