I think the python api here:
http://www.openml.org/guide/#!python

should have the pip command to install the package form github, and like 5 lines to train a scikit-learn classifier on a specific data set.
I guess we should do that after the API is solid, but I think it's very important for adoption.