We examine how the addition of an explicit MPI messaging layer to the Python language allows developers to craft flexible and powerful interfaces to high performance, massively parallel (1000+ processors) simulation codes. In designing these code, a great deal of care is given to the performance and capability of the underlying physics or mathematics of the simulation. Just as important, however, is the framework that ties the various parts of the simulation together. A framework must be both flexible and extensible. Flexible to allow components of the simulation to work together in (perhaps) unanticipated ways. Extensible to allow new components and algorithms to be quickly prototyped or added to the simulation. We use the power of the scripting language Python to form the core of our framework. Python is more complete than a handcrafted simulation control language would otherwise be. Python becomes the glue that combines otherwise unrelated components for computation, graphics, and analysis of data. We present how using pyMPI, an Open Source extension to Python, we can control and co-ordinate SPMD and MIMD simulations up to thousands of processors. We will also present PYFFLE, an Open Source extension generator, which we use to integrate C++ and FORTRAN simulation components into Python. We will also discuss initial findings from our use of PyCOT, a Python to C++ translator which we will use to accelerate user extensions written in Python.