Scientific research has been greatly advanced by the increasing power of computing systems; large scale computations on a large amount of data sets become feasible. Grid and agent computing are of great importance in the realization of such systems. On the one side, Grid computing provides large scale infrastructure that enables the sharing of resources among virtual organizations. On the other side, agent computing provides intelligent mechanisms for the coordination and cooperation of autonomous distributed software systems: the so ? called Multi - Agent Systems. Several MAS have been applied to multiple scientific, engineering and industry domains for modeling and simulating complex systems, controlling system logic and managing business processes. The problems addressed by MAS require increasing computing power and resources. Therefore, Grid is considered the appropriate infrastructure where the requirements of a MAS can be realized. We investigate the state?of?the?art developments in both MAS and Grid technologies, and we propose a framework that efficiently executes legacy MAS on the Grid.