There is a number of projects that have aims very similar to OMPC. OMPC is unique as a MATLAB(R)-to-Python syntax translator. The most similar of the listed projects is "pym". Pym compiles m-files directly to Python bytecode. This can be achieved with some effort in OMPC as well. This approach will however face a lot more issues when Python byte-code changes with different versions.
the older projects can be still used (maybe some small corrections are necessary) and they use the approach of wrapping the whole engine. This is basically embedding and comes with the problem that variables have to live and be transfered between two separate workspaces.
Another problem with the bridging projects is that they do not free the user from the commercial and expensive MATLAB(R).
I am actually very surprised that this never happened with octave. GNU Octave project provides a library called liboctave that offers possibly all the functionality of Octave for embedding. I the case of wrapping liboctave one could create a simple numpy wrapper for each Matrix object a keep all variables available for direct access by numpy.
A number of times similar projects were either discussed or initiated:
There is a number of discussions, I just need some time to find them.