Applying fault detection and correction techniques in a context become a complicated issue in an environment like computational grid. Agent technology promises a more flexible approach, easier installation and management of the agent framework, and better ability to autonomously recover from failures. Understanding and handling failures becomes additional burden for application developers in a distributed environment. So tolerating the faults and maintaining the performance of the grid becomes a major challenge in the grid computing environment. The work was aimed at building a fault tolerant computational grid using agent-based scheduler. It was found that agent technology offers many advantages in terms of ease of deployment, usage and the ability to control the scheduling of grid applications at a higher level. The framework provided a generic mechanism for incorporating agents into grid applications. It was applied to the meta-scheduler with multi criteria approach. This work also does performance evaluation of the proposed approach with respect to make span time and average scheduling time of each job.