Uv

UV - Python Package Manager Skill

Overview

UV is an extremely fast Python package and project manager written in Rust. This skill provides guidance on using UV for Python development, with particular focus on MCP (Model Context Protocol) server integration and modern tool management workflows.

UV replaces multiple tools: pip, pip-tools, pipx, poetry, pyenv, twine, virtualenv, and more - delivering 10-100x faster performance through intelligent caching and parallel operations.

Version Awareness

Recommended Version: UV 0.9.7+ (Latest as of October 2025)

Before starting, check your UV version:

uv --version

Important Version-Specific Changes:

• UV 0.9.6+: Python 3.14 is now the default (previously 3.13)
• UV 0.9.6+: Free-threaded Python 3.14+ supported without explicit opt-in
• UV 0.9.6+: uv build --clear flag available for cleaning build artifacts
• UV 0.9.7+: Security updates for tar/ZIP archive handling

If your version is older than 0.9.0, upgrade for the best experience:

# Using pip
pip install --upgrade uv

# Or reinstall using official installer
# Windows
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

# Unix/Mac
curl -LsSf https://astral.sh/uv/install.sh | sh

See Recent Changes Reference for detailed version information and migration guidance.

When to Use This Skill

Use this skill when:

• Setting up Python virtual environments and managing Python versions
• Installing and managing Python CLI tools (development tools, utilities)
• Running MCP servers with UVX
• Deciding between uv tool install vs uvx for package execution
• Configuring VS Code or other IDEs for MCP server integration
• Migrating from pip, pipx, or poetry to UV
• Troubleshooting UV-related issues

Skip this skill when:

• You need basic Python package installation only (standard pip documentation may suffice)
• Working with legacy Python 2.x projects

Core Concepts

1. UV Commands Overview

UV provides several commands for different use cases:

2. Tool vs UVX Decision Tree

Need to run a Python package?
|
├─ Use daily/frequently?
|  └─ YES → `uv tool install package`
|     Examples: black, pytest, flake8, mypy
|
├─ MCP server?
|  └─ YES → `uvx package` or `uvx --from path script.py`
|     Examples: mcp-server-sqlite, custom MCP servers
|
├─ Testing/one-off execution?
|  └─ YES → `uvx package`
|     Examples: testing new tools, version comparison
|
└─ Local development script?
   └─ YES → `uvx --from . script.py`
      Examples: project-specific scripts

3. MCP Server Execution Patterns

Published Packages (No working directory needed):

{
  "servers": {
    "sqlite": {
      "command": "uvx",
      "args": ["mcp-server-sqlite", "--db-path", "/path/to/db"]
    }
  }
}

Local Development (Use --from flag):

{
  "servers": {
    "my-server": {
      "command": "uvx",
      "args": [
        "--from", "/absolute/path/to/project",
        "server.py",
        "--config", "config.json"
      ]
    }
  }
}

Key insight: --from flag IS the working directory reference for UVX.

4. Virtual Environment Management

UV works seamlessly with Python's built-in venv:

# Create virtual environment
python -m venv .venv

# Activate (Windows Git Bash)
. .venv/Scripts/activate

# Activate (Windows CMD)
.venv\Scripts\activate.bat

# Activate (Linux/Mac)
source .venv/bin/activate

# Install packages with UV
uv pip install -r requirements.txt

Common Workflows

Development Tools Setup

# Install development tools once
uv tool install black
uv tool install flake8
uv tool install mypy
uv tool install pytest

# Use daily
black .
flake8 src/
mypy src/
pytest tests/

MCP Server Usage

# Test published MCP servers
uvx mcp-server-sqlite --db-path test.db
uvx mcp-server-git --repository /path/to/repo

# Local MCP server development
uvx --from /path/to/project server.py --env config.env

Project Initialization

# Create new project with UV
uv init my-project
cd my-project

# Add dependencies
uv add requests fastapi

# Run project
uv run python main.py

Python Version Management

# List available Python versions
uv python list

# Install default Python version (3.14 in UV 0.9.6+)
uv python install

# Install specific Python version
uv python install 3.12
uv python install 3.13

# Use in project
uv python pin 3.12

Note: As of UV 0.9.6, Python 3.14 is the default version. If you need Python 3.13 or earlier, explicitly specify the version.

Inline Script Dependencies (PEP 723)

UV supports defining dependencies directly in Python script comments:

#!/usr/bin/env -S uv run --script
# /// script
# dependencies = [
#   "requests",
#   "pandas",
# ]
# ///

import requests
import pandas as pd

# Your code here

Run with automatic dependency installation:

# UV installs dependencies automatically
uv run script.py

Benefits:

• Self-contained single-file scripts
• No pyproject.toml needed
• Easy sharing and distribution
• Perfect for utilities and automation

See Inline Script Metadata Reference for comprehensive examples including MCP servers, web applications, data processing, and CLI tools.

Integration Patterns

VS Code MCP Configuration

For .vscode/mcp.json or user settings:

{
  "servers": {
    "published-server": {
      "type": "stdio",
      "command": "uvx",
      "args": ["mcp-server-sqlite", "--db-path", "${workspaceFolder}/db.sqlite"]
    },
    "local-dev": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "--from", "${workspaceFolder}",
        "src/server.py"
      ]
    }
  }
}

Continue IDE Configuration

For .continue/config.json:

{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "transport": {
          "type": "stdio",
          "command": "uvx",
          "args": ["mcp-server-fetch"]
        }
      }
    ]
  }
}

GitHub Actions CI/CD

- name: Setup UV
  uses: astral-sh/setup-uv@v1

- name: Install dependencies
  run: uv pip install -r requirements.txt

- name: Run tests
  run: uv run pytest

Best Practices

Tool Management

DO:

• Use uv tool install for development tools used frequently
• Use uvx for MCP servers (follows community patterns)
• Keep tools isolated in their own environments
• Regularly upgrade tools with uv tool upgrade --all

DON'T:

• Use global pip for CLI tools (causes dependency conflicts)
• Install MCP servers with uv tool install (against community patterns)
• Use uvx for daily development tools (unnecessary overhead)
• Mix pip and uv tool installations

MCP Server Patterns

DO:

• Use UVX for all MCP server execution
• Use --from for local development
• Pin versions for production (package@1.2.3)
• Use environment variables for configuration

DON'T:

• Install MCP servers globally
• Mix working directory approaches
• Use @latest in production (unstable)
• Forget to specify absolute paths with --from

Virtual Environments

DO:

• Use python -m venv for project environments
• Activate before installing packages
• Use uv pip install for faster package installation
• Document activation commands in README

DON'T:

• Install packages globally
• Mix venv and system Python packages
• Forget to activate before development
• Commit .venv directory to version control