Self-Contained Python Scripts with uv

Create Python scripts that are fully self-contained — no virtualenv setup, no requirements.txt, no pyproject.toml. Just a single .py file that anyone can run with uv run.

Script Structure

Every script follows this structure:

#!/usr/bin/env -S uv run --script
# /// script
# requires-python = ">=3.14"
# dependencies = [
#   "some-package",
# ]
# ///

"""One-line description of what this script does."""

# imports and code here

Rules

1. Shebang: Always #!/usr/bin/env -S uv run --script as the first line.
2. Inline metadata: PEP 723 block immediately after the shebang.
3. requires-python: Always specify. Default to >= 3.14 unless the user specifies otherwise.
4. Maximum Python version: If a dependency is known to be incompatible with a newer Python version, add an upper bound (e.g., requires-python = ">=3.14,<3.15").
5. Dependencies: List all third-party imports in the dependencies array.

---

PEP 723 Inline Script Metadata

The metadata block uses TOML syntax, prefixed with # on every line:

# /// script
# requires-python = ">=3.14"
# dependencies = [
#   "requests>=2.31",
#   "rich",
# ]
# ///

Syntax Rules

• Opening line: exactly # /// script
• Closing line: exactly # ///
• Every interior line starts with # followed by a space (if content follows)
• Content is TOML with the # prefix stripped
• Only one script block per file

Supported Fields

Field	Type	Description
requires-python	str	Version specifier (e.g., ">=3.14")
dependencies	list[str]	PEP 508 dependency specifiers
[tool.*]	table	Tool-specific config (same as pyproject.toml)

Examples

Minimal (no dependencies):

# /// script
# requires-python = ">=3.14"
# ///

With pinned versions:

# /// script
# requires-python = ">=3.14"
# dependencies = [
#   "httpx>=0.27,<1",
#   "rich>=13",
#   "pydantic>=2,<3",
# ]
# ///

---

CLI Design

All scripts MUST have a CLI interface. The choice of CLI framework depends on complexity.

Simple Scripts: argparse

Use argparse for scripts with a flat set of arguments (no subcommands). This is the default unless the user asks for something else.

#!/usr/bin/env -S uv run --script
# /// script
# requires-python = ">=3.14"
# dependencies = []
# ///

"""Resize images in a directory."""

import argparse
from pathlib import Path


def main() -> None:
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("source", type=Path, help="Source directory")
    parser.add_argument("-o", "--output", type=Path, default=None, help="Output directory (default: source/resized)")
    parser.add_argument("-w", "--width", type=int, default=800, help="Target width in pixels (default: 800)")
    parser.add_argument("--quality", type=int, default=85, choices=range(1, 101), metavar="1-100", help="JPEG quality (default: 85)")
    parser.add_argument("-v", "--verbose", action="store_true", help="Print detailed output")
    args = parser.parse_args()

    output = args.output or args.source / "resized"
    # ... implementation ...


if __name__ == "__main__":
    main()

Complex Scripts: Cyclopts

Use Cyclopts for scripts with multiple commands or subcommand groups. Cyclopts uses Python's native type hints for argument parsing — no Argument() / Option() wrappers needed.

#!/usr/bin/env -S uv run --script
# /// script
# requires-python = ">=3.14"
# dependencies = [
#   "cyclopts>=4",
# ]
# ///

"""Project scaffolding tool."""

from pathlib import Path
from typing import Annotated, Literal

from cyclopts import App, Parameter, validators

app = App(name="scaffold", help=__doc__, version="1.0.0")


@app.command
def init(
    path: Path = Path("."),
    template: Literal["default", "minimal", "full"] = "default",
    *,
    force: bool = False,
):
    """Initialize a new project.

    Parameters
    ----------
    path
        Directory to initialize.
    template
        Project template to use.
    force
        Overwrite existing files.
    """
    print(f"Initializing {template!r} project at {path}")


# ── Subcommand group ──────────────────────────────────────

db = app.command(App(name="db", help="Database operations."))


@db.command
def migrate(*, dry_run: bool = False):
    """Run pending migrations.

    Parameters
    ----------
    dry_run
        Show SQL without executing.
    """
    print(f"{'[DRY RUN] ' if dry_run else ''}Running migrations...")


@db.command
def seed(
    count: Annotated[int, Parameter(validator=validators.Number(gte=1))] = 10,
    *,
    table: str | None = None,
):
    """Seed the database with test data.

    Parameters
    ----------
    count
        Number of records to insert.
    table
        Specific table to seed (all if omitted).
    """
    target = f"table {table!r}" if table else "all tables"
    print(f"Seeding {count} records into {target}")


if __name__ == "__main__":
    app()

Cyclopts Key Patterns

from cyclopts import App, Parameter, validators
from typing import Annotated, Literal

app = App(name="tool", help="...", version="1.0.0")

# Default action (no subcommand given)
@app.default
def main(): app.help_print()

# Command (name auto-derived: foo_bar -> foo-bar)
@app.command
def foo_bar(): ...

# Subcommand group
sub = app.command(App(name="sub", help="..."))
@sub.command
def action(): ...

# Positional-only (no --flag generated)
def cmd(src: Path, dst: Path, /): ...

# Keyword-only (must use --flag)
def cmd(*, verbose: bool = False): ...

# Short alias
output: Annotated[str, Parameter(alias="-o")] = "out"

# Validator
port: Annotated[int, Parameter(validator=validators.Number(gte=1, lte=65535))] = 8080

# Choices
fmt: Literal["json", "yaml", "toml"] = "json"

# Entry point
if __name__ == "__main__":
    app()

Cyclopts vs Typer: Cyclopts uses Python's / and * markers for positional vs keyword. No Argument() / Option() needed. Decorator parentheses are optional: @app.command works without ().

---

Rich Output Formatting

Use Rich when the script benefits from formatted terminal output (tables, panels, syntax highlighting, styled text).

# dependencies = ["rich"]

from rich.console import Console
from rich.table import Table
from rich import print  # drop-in replacement for print()

console = Console()
console.print("[bold green]Success![/] Operation completed.")

table = Table(title="Results")
table.add_column("Name", style="cyan")
table.add_column("Status", style="green")
table.add_row("item-1", "OK")
console.print(table)

Only add rich as a dependency when formatted output provides clear value. Plain print() is fine for simple scripts.

---

Progress Bars: alive-progress

Use alive-progress when a script performs a loop over many items or a long-running operation. Do NOT use a progress bar for tasks expected to complete in under 2 seconds.

# dependencies = ["alive-progress"]
from alive_progress import alive_bar

Basic Usage

with alive_bar(len(items), title="Processing") as bar:
    for item in items:
        process(item)
        bar()

Full-Featured Example

with alive_bar(
    len(files),
    title="Uploading",
    unit="files",
    dual_line=True,
) as bar:
    for f in files:
        bar.text = f"-> {f.name}"
        upload(f)
        bar()

Key Parameters

Parameter	Type	Default	Use When
total	int | None	None	Always specify when the count is known. None = unknown/streaming mode (no ETA).
title	str | None	None	Always provide a short, descriptive label (shown left of bar).
unit	str	""	Items have a natural unit ("files", "rows", "req", "B").
scale	str | None	None	Byte-like units: "SI" (1000-based), "IEC" (1024-based, KiB/MiB/GiB).
dual_line	bool	False	bar.text messages are long and would clutter the bar line.
spinner	str | None	from theme	Override spinner style. Named strings: classic, dots, waves, pulse, etc.
bar	str | None	from theme	Override bar fill style: smooth, classic, blocks, bubbles, etc.
theme	str	"smooth"	Preset bundle of spinner + bar + unknown style. Options: smooth, classic, scuba, musical.
manual	bool	False	You know the percentage but not item count. Call bar(0.0-1.0).
force_tty	bool | None	None	True for PyCharm/Jupyter. False for CI (receipt only).
receipt_text	bool	False	True to show the last bar.text in the final summary line.

The bar Handle

with alive_bar(total, title="Work") as bar:
    bar()                        # advance by 1
    bar(5)                       # advance by 5
    bar.text = "current status"  # situational message (inline or second line with dual_line)
    bar.title = "Phase 2"        # update the left-side title mid-run
    bar.current                  # read current count

Operating Modes

total	manual	Mode	bar() call
provided	False	Auto (default)	bar() increments by 1
None	False	Unknown	bar() increments (no ETA, animated)
provided	True	Manual	bar(0.0-1.0) sets percentage

Iterator Shortcut

from alive_progress import alive_it

for item in alive_it(items, title="Processing"):
    process(item)

Common Patterns

# ── Bytes with IEC scaling ─────────────────────────────────
with alive_bar(file_size, unit="B", scale="IEC", title="Download") as bar:
    for chunk in stream:
        write(chunk)
        bar(len(chunk))

# ── Unknown total (streaming) ──────────────────────────────
with alive_bar(title="Reading stream") as bar:
    for record in stream:
        process(record)
        bar()

# ── Manual percentage ──────────────────────────────────────
steps = ["fetch", "transform", "load"]
with alive_bar(manual=True, title="Pipeline") as bar:
    for i, step in enumerate(steps):
        bar.text = step
        run(step)
        bar((i + 1) / len(steps))

Gotchas

• No nesting: Do not nest with alive_bar() blocks. Use sequential bars instead.
• total must be int: Not float. Cast if needed: alive_bar(int(total)).
• bar() outside the with block is silently ignored.
• print() inside the with block works correctly — alive-progress hooks stdout and renders print output above the bar.
• bar.text is not shown in the final receipt unless receipt_text=True.

---

Interactive Prompts: questionary

Use questionary when the script needs user input beyond simple CLI arguments. ALWAYS provide a CLI argument alternative so the script can run non-interactively (e.g., in CI).

# dependencies = ["questionary"]
import questionary

Pattern: CLI Args with Interactive Fallback

def get_config(args: argparse.Namespace) -> dict:
    """Resolve config from CLI args, falling back to interactive prompts."""
    name = args.name or questionary.text("Project name?", default="my-project").ask()
    template = args.template or questionary.select(
        "Template?",
        choices=["default", "minimal", "full"],
    ).ask()
    return {"name": name, "template": template}

Prompt Types

text — free-form input

name = questionary.text("Your name?", default="World").ask()

password — masked input

secret = questionary.password("API key?").ask()

confirm — yes/no

proceed = questionary.confirm("Continue?", default=True).ask()

select — pick one (arrow keys)

choice = questionary.select(
    "Environment?",
    choices=["development", "staging", "production"],
    default="development",
).ask()

checkbox — pick many (Space to toggle)

selected = questionary.checkbox(
    "Features to enable?",
    choices=["auth", "logging", "metrics", "tracing"],
).ask()

path — file/directory with Tab completion

config = questionary.path("Config file?", default="./config.yaml").ask()

autocomplete — text with suggestions

lang = questionary.autocomplete(
    "Language?",
    choices=["Python", "Rust", "Go", "TypeScript", "Java"],
).ask()

Choice Objects

from questionary import Choice, Separator

choices = [
    Choice("Production", value="prod"),
    Choice("Staging", value="staging"),
    Separator("--- Dev ---"),
    Choice("Local", value="local"),
    Choice("Docker", value="docker", disabled="Not available"),
]

Validation

questionary.text(
    "Port?",
    validate=lambda v: True if v.isdigit() and 1 <= int(v) <= 65535 else "Must be 1-65535",
).ask()

Return Value

All .ask() calls return None if the user cancels with Ctrl-C. Always handle this:

name = questionary.text("Name?").ask()
if name is None:
    print("Cancelled.")
    raise SystemExit(1)

---

Fuzzy Selection: iterfzf

Use iterfzf when the user needs to select from a large list with fuzzy search. fzf is bundled in the package — no separate install needed.

# dependencies = ["iterfzf"]
from iterfzf import iterfzf

Basic Usage

# Single selection
choice = iterfzf(["apple", "banana", "cherry", "date"])

# Multi-selection (Tab to toggle)
selected = iterfzf(
    ["alpha", "beta", "gamma", "delta"],
    multi=True,
    prompt="Pick items > ",
)

Key Parameters

Parameter	Default	Description
multi	False	True = multi-select with Tab. Returns list[str].
prompt	" >"	Prompt string shown in fzf UI.
query	""	Pre-filled search query.
exact	False	True = exact substring match instead of fuzzy.
case_sensitive	None	True/False/None (smart-case).
preview	None	Shell command for preview pane (e.g., "cat {}").
header	None	Sticky header text below the prompt.
ansi	None	True to render ANSI colors in items.

Return Types

multi	User selects	User cancels (Esc)
False	str	None
True	list[str]	None

Ctrl-C raises KeyboardInterrupt.

Feeding Large / Lazy Data

iterfzf accepts any iterable — items are streamed lazily:

import subprocess

# Stream git log lazily
def git_commits():
    proc = subprocess.Popen(["git", "log", "--oneline", "-100"], stdout=subprocess.PIPE, text=True)
    yield from (line.strip() for line in proc.stdout)

commit = iterfzf(git_commits(), prompt="Pick commit > ")

---

Complete Example: Full-Featured Script

#!/usr/bin/env -S uv run --script
# /// script
# requires-python = ">=3.14"
# dependencies = [
#   "httpx>=0.27",
#   "rich>=13",
#   "alive-progress>=3",
# ]
# ///

"""Fetch and display GitHub repository statistics."""

import argparse
from pathlib import Path

import httpx
from alive_progress import alive_bar
from rich.console import Console
from rich.table import Table

console = Console()


def main() -> None:
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("repos", nargs="+", help="GitHub repos (owner/name)")
    parser.add_argument("-o", "--output", type=Path, default=None, help="Save results to JSON file")
    parser.add_argument("--token", default=None, help="GitHub API token")
    args = parser.parse_args()

    results = []
    with alive_bar(len(args.repos), title="Fetching repos", unit="repos") as bar:
        for repo in args.repos:
            bar.text = f"-> {repo}"
            resp = httpx.get(
                f"https://api.github.com/repos/{repo}",
                headers={"Authorization": f"Bearer {args.token}"} if args.token else {},
            )
            resp.raise_for_status()
            results.append(resp.json())
            bar()

    table = Table(title="Repository Stats")
    table.add_column("Repository", style="cyan")
    table.add_column("Stars", justify="right", style="yellow")
    table.add_column("Forks", justify="right")
    table.add_column("Language", style="green")

    for r in results:
        table.add_row(r["full_name"], str(r["stargazers_count"]), str(r["forks_count"]), r.get("language", "—"))

    console.print(table)

    if args.output:
        import json

        args.output.write_text(json.dumps(results, indent=2))
        console.print(f"[dim]Saved to {args.output}[/dim]")


if __name__ == "__main__":
    main()