Awesome Python Dev

Curated AI development toolkits, ready for your coding agents.

English · 简体中文

Browse the website · Getting started · Website maintenance

Find useful, established GitHub projects for AI assisted development, then install a selected toolkit into your project with one command. Skills supply reusable workflows, rules activate engineering and security guidance, MCP servers connect tools and context, and an LSP bridge exposes code intelligence to your agent.

The installer is written in Python, with a dedicated Python toolkit. The collection also serves web and general software development across Claude Code, Codex, Cursor, VS Code Copilot, and OpenCode.

The v0.3.0 catalog includes 54 entries: 30 installable resources (20 skills, 6 rule packs, 3 MCP servers, and Serena's LSP bridge) and 24 references (7 native language servers and 17 other tools). Choose from 11 profiles across 5 coding agents.

Contents

• Quick start
• npm shell and runtime updates
• Explore the collection
• Choose your toolkit
• Supported agents
• Rules and framework guidance
• LSP and security tools
• Everyday commands
• How installation works
• What makes a project a good fit
• Contribute
• Acknowledgments

Quick start

Install uv, then run this inside the project you want to equip:

uvx awesome-python-dev init --agents codex,claude --profile starter

Or use npm:

npx awesome-python-dev init --agents codex,claude --profile starter

The Python package requires Python 3.11+. The npm launcher requires Node.js 22.14+ and Python 3.11+ or an existing uv installation. When Python is unavailable, uv can provision Python 3.13. Both entry points run the same toolkit.

This downloads the selected skills and common rules, then writes project-local agent configuration. Restart or reload your agent, review its MCP approval prompts, and start working. Codex loads project configuration only for a trusted project; see the agent guide.

For repeated use, install the CLI once:

uv tool install awesome-python-dev
apd list
apd init --agents codex,claude --profile starter --dry-run
apd init --agents codex,claude --profile starter
apd doctor

For a global npm installation, use npm install --global awesome-python-dev instead. Both distributions provide apd and awesome-python-dev; choose one global installation method. Refresh the Python installation with uv tool upgrade awesome-python-dev, or request a particular release with uvx awesome-python-dev@0.3.0.

The registry commands do not require a Git checkout. A version-pinned source fallback is also available when Git is installed:

uvx --from 'git+https://github.com/m9rco/awesome-python-dev.git@v0.3.0' apd init --agents codex,claude --profile starter

MCP runtime prerequisites depend on your selection: npm-based servers need Node.js and npx; Python-based servers need uvx. Your coding agent starts these processes when it loads the configuration. apd init configures MCP servers; it does not launch them or install your coding agent.

The starter profile needs Node.js 20.18.1 or newer for Context7. On Windows, npm recipes use cmd /c npx. Existing JSONC configuration is left untouched: convert it to JSON before adding MCP entries, or install skills only with --tools.

Example preview from apd init --agents codex --profile starter --dry-run:

Preview only (no downloads or writes):
codex: skill systematic-debugging -> .agents/skills/systematic-debugging
codex: skill test-driven-development -> .agents/skills/test-driven-development
codex: skill verification-before-completion -> .agents/skills/verification-before-completion
codex: MCP context7 -> .codex/config.toml
codex: rules ecc-common-rules -> native project instructions

npm shell and runtime updates

The npm shell and Python runtime have independent versions. The shell is a small entry point; it downloads the bundled runtime from GitHub Releases on first use, checks its SHA-256 hash, and caches it. Normal commands check for a compatible runtime update at most once every 24 hours. Catalog and installer updates therefore ship without another npm upload.

apd --shell-version     # npm shell version
apd --version           # selected runtime version
apd runtime status      # inspect local selection; no network request
apd runtime update      # check the stable runtime channel now

These runtime commands belong to the npm launcher. For reproducible use, set APD_RUNTIME_VERSION=0.3.0; APD_OFFLINE=1 uses an already cached runtime. Offline mode covers the launcher, so installing uncached upstream skills can still require network access. See runtime selection and caching and the release workflow.

Explore the collection

Browse the curated catalog for project descriptions, upstream links, licenses, review dates, and installation support. The research notes record selection evidence and pinned versions.

Also explore OpenSpec, Spec Kit, GitHub MCP, uv, Ruff, and security scanners in the full reference collection.

Entries marked installable have a recipe in this release. Entries without an install recipe are references, with installation managed by their upstream projects. lsp includes both the installable Serena bridge and standalone language-server references.

apd list --kind skill
apd list --kind rules
apd list --kind mcp
apd list --kind lsp
apd list --kind reference
apd list --json

Choose your toolkit

apd profiles shows the exact membership of each profile. Rule dependencies are included automatically: language packs require common rules, and FastAPI rules also require Python rules. You can select specific installable IDs instead:

apd profiles
apd init --profile fastapi --agents codex,claude
apd init --agents codex --tools systematic-debugging
apd init --agents codex --tools serena
apd init --agents all --profile starter --dry-run

Choose the agents you actually use. all targets every supported adapter, even if the corresponding application is not installed.

Select the profile for your actual stack. Framework guidance must be checked against your project's versions; the selected Next.js/Turbopack skill targets Next.js 16+. Profiles are additive, so installing unrelated language packs accumulates instructions and may reach an agent's size limit.

Supported agents

See paths, official documentation, and compatibility notes. Adapter tests cover configuration formats and merge behavior. They do not establish end-to-end validation inside every agent application or compatibility with every historical version. The vscode target refers to Copilot in VS Code, not the separate GitHub Copilot CLI.

Rules and framework guidance

The selected ECC skills and rules are MIT licensed, pinned to a reviewed commit, and archive-hash verified. The ECC migration guide records the source, exact pins, file allowlists, dependencies, and adaptations. Original rule files are preserved under .awesome-python-dev/rules/<id>/ with attribution; adapted copies activate through each agent's instruction system.

Existing text outside the AGENTS.md block is preserved. The installer rejects conflicts and offers no force-overwrite mode. With Codex selected, it guards the generated AGENTS.md against the 32 KiB default instruction limit. It refuses oversized output rather than truncating rules.

These are selected guidance files, not the entire ECC plugin. Installation does not register ECC agents, hooks, model preferences, or background services. References to workflow roles are adapted for the host agent. Security checks, tests, migrations, and deployment still use the target project's real tools and permissions.

LSP and security tools

apd init --agents codex --tools serena configures a pinned Serena MCP launch command. It selects Python 3.13 and the LSP backend; the agent starts it later. The recipe passed MCP initialization and discovery of 22 tools in a temporary Python project. That check did not execute symbol operations or validate every agent interface. See the LSP guide for prerequisites, language-server references, OpenCode's separate native LSP support, and known limitations.

The security profile installs review guidance and rules. Semgrep, Bandit, pip-audit, and Gitleaks are catalog references for project checks and CI. Installing guidance does not install these scanners or certify a project as secure.

Everyday commands

# Equip another project without changing directories.
apd init --path ../my-project --agents codex --profile python

# Show supported agents and their project paths.
apd agents

# Check installed files, configuration, and runtime prerequisites.
apd doctor --path ../my-project

# Preview removal, then remove the installation managed by apd.
apd remove --path ../my-project --dry-run
apd remove --path ../my-project

Run apd --help or apd init --help for the complete command reference. Re-running init adds selections; choosing a different profile does not remove previously installed tools. remove removes all unchanged resources owned by apd in the project, preserving unrelated user content. Conflicting user changes are protected. Review git diff after installation, and commit the generated project configuration if your team should share it.

Upgrading from v0.1: the v0.2 CLI reads manifest schemas 1 and 2, and upgrades schema 1 during init. Keep the manifest and backups. Once upgraded, use v0.2 or newer to manage that installation; v0.1 cannot manage schema 2 manifests.

How installation works

Reviewed catalog → toolkit selection → agent adapters → project files
                                           ↓
                            tracked installation and doctor checks

• Skill and rule sources are pinned to Git commits and verified against archive SHA-256 hashes. Their selected supporting files and upstream licenses travel with the installation.
• MCP recipes use explicit package versions; the Serena bridge pins a Git commit. Recipes are translated into each agent's configuration format.
• Installation records ownership so removal can preserve unrelated settings. Modified managed files are protected from silent overwrites.
• Keep .awesome-python-dev/manifest.json for subsequent refreshes and removal. Configuration backups stay under .awesome-python-dev/backups/, with a Git ignore rule; keep them private.
• Preview with --dry-run before writing. No home-directory agent configuration is changed.
• The installer does not run downloaded skill scripts. Skills and MCP servers can guide or execute actions later through your agent, under that agent's permissions.

Version pins make changes reviewable. They do not freeze all transitive dependencies or replace reviewing upstream code. Runtime authentication, API limits, browser downloads, and client approval remain specific to the selected tool. See security and reporting.

What makes a project a good fit

We prefer tools with a concrete development use case, useful documentation, an identifiable upstream, a clear license, and evidence of continued maintenance. Stars are one discovery signal; they are not a quality guarantee or a ranking system.

An installable entry additionally needs a reproducible recipe, a reviewable source or package version, verified paths and runtime requirements, and a way to preserve upstream attribution. Rule packs also need explicit file selection, dependencies, and tested native activation. Projects that need credentials, manual setup, or a richer adapter can remain useful reference entries while an integration is evaluated.

Contribute

Suggest a tool, improve an installation recipe, or add a tested agent adapter. Start with the contribution guide and roadmap. Report installation problems using the issue template, with secrets removed from logs and configuration.

See the CI workflow and latest runs for automated checks, and the contribution guide to run them locally.

Acknowledgments

The organization follows the focused categories and contribution discipline of Awesome, the use-case-oriented curation of Awesome Python, and the MCP discovery work of Awesome MCP Servers. The project-local initialization experience is inspired by OpenSpec.

This is an independent project. Upstream authors own their tools and trademarks. Our installer and original documentation are MIT licensed; included upstream material retains its own license.