If you wish to perform VTune profiling on a single node, that is easy enough. If you don't care to execute a script, the application to be run (for a suitable version of MPI) is your mpirun command, and the command line is everything you would enter following mpirun on the command line. Absolute paths are most likely to work. You would compile any parts of the application which you want to designate as modules of interest (including MPI itself) with -g.
I suppose you could install and start VTune collector on all nodes of the cluster, then collect and analyze the behavior of your application individually on each node. I don't know from your question whether that would be relevant.
If you are attempting to profile the behavior on many nodes, including messages passed, you may be looking for the Trace Collector and Analyzer component of Intel Cluster Toolkit, which supports primarily Intel MPI and mpich-2.
