Python supports inline metadata for scripts (a feature added in 2024). This makes it possible to run standalone scripts with dependencies and Python versions managed automatically.

Many tools support this workflow, including PDM, Hatch, and {term}uv. In this tutorial, we focus on Hatch and UV. The same metadata format can also be used with other tools.

:::{seealso}

• Hatch: How to run Python scripts
• uv: Running scripts :::

Sometimes you want to share or run a single script without creating a full {term}Python package. To do this, you can use inline script metadata. This format lets you specify dependencies and Python versions at the top of your script in a comment block.

When you add metadata at the top of a script, Hatch (or PDM or uv) will use that metadata to:

• Create an isolated virtual Python environment for that script.
• Install the {term}Dependencies listed in the script into that environment.
• Use the required Python version that you specify in the metadata.

This approach is useful for workflows that you want to make reproducible, but that do not need to become full Python packages.

Inline metadata helps you make scripts reproducible. Anyone can run your script without manually creating a new environment or guessing which dependencies it needs. Hatch takes care of installing dependencies and using the correct Python version.

You will use Hatch in this example, but you can also use uv if that is your preferred tool.

First, create a new file named script.py with the block below at the top. The metadata block starts with # //github.com/ script and ends with # //github.com/. Everything in between must be {term}TOML metadata written as comments.

In the example below, the script requires Python 3.11 or newer, and NumPy is declared as a dependency.

# //github.com/ script
# requires-python = ">=3.11"
# dependencies = [
#   "numpy",
# ]
# //github.com/

import numpy as np

temperatures_c = np.array([12.4, 13.1, 14.8, 15.2, 16.0, 14.4, 13.7])
mean_temp = np.mean(temperatures_c)
anomalies = temperatures_c - mean_temp

print(f"Mean temperature: {mean_temp:.2f} C")
print("Temperature anomalies:", np.round(anomalies, 2))

Open your terminal and change to the directory where script.py lives. Then run:

hatch run script.py

On first run, Hatch will create an environment and install dependencies. On later runs, Hatch will reuse that environment so startup is faster.

:::{note} The environment name is based on the script path. If you move the script to a new location, Hatch will treat it as a new script environment. :::

You can control script-specific Hatch behavior in the same metadata block. For example, to use pip instead of uv as the installer:

# //github.com/ script
# requires-python = ">=3.11"
# dependencies = ["numpy"]
#
# [tool.hatch]
# installer = "pip"
# //github.com/

:::{admonition} Run the same script with uv

If you prefer uv, you can run the same inline-metadata script with:

uv run script.py

The same # //github.com/ script metadata block works with uv, including requires-python and dependencies.

For more on using uv to run scripts, see the guide: Running scripts with uv. :::

You may be wondering when to use scripts versus creating a package.

This depends on your use case. Scripts are often useful when:

• You have one small task, or a specific workflow that is not generalizable.
• Your workflow is still evolving, but you want to run it in a reproducible environment.
• You want reproducible dependencies quickly.
• You are sharing a single file with collaborators.

Create a full package when:

• You are building reusable modules for multiple projects.
• You need tests, documentation, releases, and long-term maintenance.
• Your codebase is growing beyond one or two scripts.

:::{seealso}

• Get to know Hatch
• Create a Python package
• Command line reference guide :::