Good idea, but I'm not sure this is an ideal approach...

1. At this stage of development, 99% of people running this are going to be using the git version anyway, in which case updating is a simple git pull + relaunch. Once the game is meaningful to users who wouldn't use git directly, we'd probably produce packaged Releases (including pyglet/dependencies), possibly as a self-contained .exe. Travis/Appveyer can automate this.
2. The sorts of users who can't be served by a git pull likely don't have git installed to be able to run git clone. Replacing the git clone with something like (untested)

r = urllib.request.urlopen('https://github.com/boskee/Minecraft/archive/master.zip')
with zipfile.ZipFile(BytesIO(r.read())) as z:
    z.extractall()

is more portable (non-Windows), is much more efficient/fast than a git clone, doesn't involve temporary files, and avoids run() in favour of pure python.
3. You still need to manually relaunch the game, hot-reloading would be sweet for development (and useful for this)... you might also be able to get away with replacing python main.py with python update_and_run_main.py which checks for + updates first, prior to import'ing/exec'ing the rest of the game, though that would prevent you from having a nice pyglet Update button.

That's fixed...

Replacing the git clone with something like (untested)

r = urllib.request.urlopen('https://github.com/boskee/Minecraft/archive/master.zip')
with zipfile.ZipFile(BytesIO(r.read())) as z:
    z.extractall()

That works, but there's a catch: It updates update.py.

So, we need to use temporary files so that while updating, the updater does not update itself(the updater overwrites the file it is in).

Oh, in the imports the urllif.request needs to be changed to urllib.request.
Also, append this after all the imports:from io import BytesIO