When building a Python package, the most common way of exposing a command line interface (CLI) is to use the “scripts” keyword in setup.py.

# file: setup.py
from setuptools import setup

setup(
  name='my_package',
  scripts=['scripts/my_cli'],
  ...
)

However, entry points represents an improved and cross-platform approach to accomplish the same goal.

# file: setup.py
from setuptools import setup

setup(
  name='my_package',
  entry_points={
    'console_scripts': [
      'my_cli': 'my_package.cli:main'
    ]
  },
  ...
)

What’s happens when the user installs the package is that pip will make sure that my_cli is available as an executable on the system. Once invoked, the main function of the my_package.cli module will be called.

$ pip install my_package
$ my_cli  # the main function is called
Usage: my_cli [OPTIONS] COMMAND [ARGS]...

The problem with the former approach is that it requires separating CLI related code into a isolated file outside the main package. With entry points it’s possible to unify all source code under the same package directory. This makes testing significantly easier.

---

There’s actually a lot more you can use entry points for. It is ideally suited to act as a simple way for third party plugins to hook into your application. A great example of such integration is employed by pytest for their excellent plugin architecture.

You can also find out more in the setuptools docs.