All the scientists I know on Python (N > 30) use matplotlib and (numpy or scipy or pandas). Some niche libraries for certain data formats are also necessary. These libraries are compatible in 2 and 3, and are easy to install with pip or conda, so everyone I've shown them to has made the jump quickly from Matlab, IDL, and others.

Web people Django or similar and requests to make http... then a database driver. that gets you pretty far.

Numpy undergirds the scientific computing stack in Python.