Workflow scheduling is the way to schedule the complex scientific workflow. In a cloud platform, the resources can be utilized more effectively and efficiently by scheduling the workflow properly. In this paper, an agentbased workflow scheduling mechanism on cloud environment is proposed. We implement a set of agents that can federate multiple clouds and automatically monitor the entire system, and manage the job if it's a workflow job. We also implement a dynamic job dispatching algorithm to reduce the VM usage amount and improve the resource utilization.