Grids collect the geographically distributed heterogeneous resources and use their aggregated power to solve large-scale scientific problems. In order to aggregate scientific computing libraries and parallel programming environments, a Grid computing agent is introduced. The agent acts as the middleware of grid computing. It coordinates several parallel programming environments to support the heterogeneous parallel computing and integrates scientific computing libraries to deliver the efficient solution of large problems. The agent consists of four modules: (1) parallel compilation module for compiling, debugging, and distributing codes. (2) parallel execution module for resource scheduling, application program execution and result redirection. (3) scientific computing module as a graphical user interface which is implemented by integrating Matlab and NetSolve [1]. (4) resource management module providing a single system image for the whole system. Based on the agent, a web portal is developed which provides a uniform service entry for end users. This approach improves the system usability significantly.