Breaking this down into at least three(four) different parts:

1. establish the properties of the current inline theme (e.g., transparent background, fig size)
2. Take properties and make into matplotlib theme
3. Use entrypoints to register theme with matplotlib (if possible w/o using setuptools)
4. have %matplotlib inline use the theme by default

Ok, so 0. can be found in this commit in ipykernel

as of right now its

rc = Dict({'figure.figsize': (6.0,4.0),
        # play nicely with white background in the Qt and notebook frontend
        'figure.facecolor': (1,1,1,0),
        'figure.edgecolor': (1,1,1,0),
        # 12pt labels get cutoff on 6x4 logplots, so use 10pt.
        'font.size': 10,
        # 72 dpi matches SVG/qtconsole
        # this only affects PNG export, as SVG has no dpi setting
        'savefig.dpi': 72,
        # 10pt still needs a little more room on the xlabel:
        'figure.subplot.bottom' : .125
        },
        help="""Subset of matplotlib rcParams that should be different for the
        inline backend."""
    ).tag(config=True)

Next is to figure out the matplotlib theme syntax

I suggest that you you do 0 and 1 via by implementing matplotlib/matplotlib#6738 and then using it from inside an ipython notebook with mpl 1.5

Or just start from https://github.com/matplotlib/matplotlib/blob/master/lib/matplotlib/mpl-data/stylelib/classic.mplstyle

You can just add a path to https://github.com/matplotlib/matplotlib/blob/master/lib/matplotlib/style/core.py#L33 mpl should probably grow a nicer API for managing that list.

This has been bumped through several minor release milestones now. Is anyone intending to work on it, or shall we wishlist it?

Wishlist'ed

Just a little update: the RC overrides have been removed in ipython/matplotlib-inline#14

From today's POV, would it still make sense to create an "inline" theme?

My thought is no - that we should finish fixing any small defaults deviations we might have there (looking at the dpi issue you mentioned) but otherwise let users access the normal matplotlib mechanisms as much as possible.

I think we should close this, but will wait a bit in case there's a difference of opinion.