With the growing requirement of computational power and hardware resources, Grid has become the major role in several researches and business areas. With the advanced capabilities of mobile, lightweight, and thin devices, Grid computing has, recently, migrated to pervasive and utility computing. To address these problems, we designed and implemented an agent-based lightweight Grid application toolkit, which is called MAGE (Mobile Agent-based Grid Environment). It consists of several hierarchical layers with dynamically controllable agents for archiving isolation of dependency among communication, message interpretation, and actual jobs. Because an agent represents a service in MAGE, each element of application can be replaced with a new task without intervention of other tasks. In this paper, we will describe the design issues and advantages in using our system.