Nolan,

Thanks for the pointer to pyxnat!  That's exactly the kind of thing I'm looking for.

I'm thinking of creating a layer above all this that does all the interop with datasets under the covers.  It would use a text-based config file for credentials, would throw exceptions to tell users any manual config processes are necessary for any particular data source (like user registration), and would expose metadata, data files and downloading through a common interface, and would work well with nipy.

The data sources that I've seen so far are:
* ABIDE
* HCP (Human Connectome Project)
* NDAR
* NITRC
* OASIS
* OpenfMRI
* XNAT

Interested to hear any others.  Creating plugins to access data / metadata through a script interface seems a requirement for any real automated data processing to any of these sites, and putting the extra wrapper layer to expose them in a common interface seems really easy.

Glad to hear any thoughts about it!  I'm going to try and head in that direction as I figure things out for myself, at least...
https://github.com/bcipolli/pyopenmri

Ben