In this paper, we discuss how to shape a MAS infrastructure to support an agent-oriented role-based access control model (RBAC-MAS). First, we introduce the RBAC model, and show how it can be extended to capture the essential features of agent systems. Then, we extrapolate the core requirements of an infrastructure for RBAC-MAS, and depict a possible approach based on ACCs (agent coordination contexts). The conceptual framework for a RBAC-MAS infrastructure exploiting ACCs is subsequently formalised through a process-algebraic description of the main infrastructure entities: this is meant to serve as a formal specification of both the infrastructure and the language for expressing roles, operations and policies in RBAC-MAS.