Introducing cctk: the Claude Code Toolkit | Claude Code Toolkit
Announcements

Introducing cctk: the Claude Code Toolkit

A community-vetted collection of slash commands, subagents, hooks, and MCP integrations. One install, sensible defaults, deeply customizable.

MK
Mubashir K. August 5, 2026 · 8 min read

Every team that starts using Claude Code goes through the same journey. Week one is "wow this is powerful." Week three is "I should build some slash commands for the things I do a lot." Month two is "I really should build these subagents properly, but I haven't had time." Month six is "we've got a folder of half-finished commands and no one owns them."

cctk — the Claude Code Toolkit — exists so your team can skip that whole arc. It's a community-vetted collection of the things every team eventually needs to build: slash commands, subagents, lifecycle hooks, and MCP integrations. Installed with one command, sensible by default, fully customizable if your team has strong opinions.

Today we're launching cctk v1.5.2 publicly. This post walks through what's in the box, why we built it, and how to get started in the next five minutes.

The tools that work well for AI-augmented dev aren't what most teams end up building on their own. They're what small groups of teams building carefully over time converge on.
— cctk contributor

Why we built cctk

The obvious question: why not just have every team build their own? The answer comes from watching a few dozen teams over the past year. Everyone builds versions of the same commands. Everyone iterates through the same mistakes. Everyone's week-one slash commands are bad, and everyone's month-three ones are much better — but almost identical to what other teams converged on independently.

Commands need iteration to be good. A first-draft /test-unit command asks Claude to "write tests for this file." A third-iteration command specifies exactly what coverage to hit, what conventions to match, what to skip. The difference in output quality is enormous, and getting there means writing bad versions and noticing what breaks.

cctk skips the iteration cost. Every command in the library went through 3-6 revisions before shipping. Real teams used them. The prompts have been sanded down by contact with real code. That's the value — not the raw commands (you could write those yourself in an hour), but the versions that have been through the fire.

What's in the box

The v1.5.2 release includes four categories of tools:

  • 85 slash commands across 21 categories — testing, refactoring, docs, debugging, deployment, frontend, backend, security, git, database, monorepo, styling, devops, AI, performance, API design, forms & state, mobile, workers & queues, CLI tools, i18n.
  • 12 subagents for delegated work — research-agent, pr-reviewer, refactor-agent, incident-investigator, security-researcher, docs-agent, test-writer, modernizer, onboarding-agent, debug-agent, code-review-agent, architecture-reviewer.
  • 12 lifecycle hooks for team standards — block-secrets, block-dangerous-commands, auto-format, pre-commit-guard, path-guard, commit-message-guide, tool-audit, cost-tracker, session-logger, and more.
  • 12 MCP integrations for external services — GitHub, Sentry, Datadog, Linear, Slack, Notion, Figma, Stripe, PostgreSQL, AWS, Cloudflare, and Grafana.

Everything is versioned, tested, and updated as Claude Code itself evolves. New tools go in every week; the community-driven process means what ships is what teams actually use, not what looks good in a demo.

Installing cctk in one minute

The install is designed to be forgettable — you run one command, cctk sets up .claude/ in your project with sensible defaults, and you keep working.

Install and initialize
npx cctk@latest init

# What happens:
# 1. Detects your stack (Next.js, Django, Go, etc.)
# 2. Creates .claude/ with the right defaults
# 3. Installs 10-15 commands appropriate for your stack
# 4. Adds one starter subagent (research-agent)
# 5. Adds two starter hooks (block-secrets, auto-format)
# 6. Prints next steps

That's intentionally minimal. cctk installs enough to be immediately useful and nothing else. Everything else is opt-in: browse the catalog, install what you need. We've seen too many "install everything" experiences that overwhelm new users; this one's built for gradual adoption.

Adding more from the catalog
cctk add /test-unit /test-integration    # Add test commands
cctk add --agent pr-reviewer              # Add a subagent
cctk add --hook block-secrets             # Add a hook
cctk add --mcp sentry                     # Add an MCP integration

cctk list                                 # See what's installed
cctk update                               # Update everything to latest
Pro tip

Run cctk add --category=testing to install all commands from a category at once. Useful when starting a new project and you want the full testing suite ready to go.

Design principles

A few design principles that guide what goes into cctk (and what stays out):

  1. Match your existing conventions. Commands read your existing code before generating anything. If your tests use Vitest, generated tests use Vitest. If your endpoints return {data: T, meta: {...}}, new endpoints do too. No "here's a generic template" output.
  2. Team-shareable by design. Everything is version-controlled in .claude/, so it travels with the repo. New team members clone and have the same commands automatically. No "install this on your machine" step.
  3. Customization over configuration. Every command has an obvious file you can edit. Don't like the prompt? Change it. Want to swap the model? Change the frontmatter. cctk's versions are starting points, not black boxes.
  4. Editor-agnostic. Claude Code is a CLI tool that works with any editor. cctk doesn't require you to switch to a specific IDE. Vim, JetBrains, Zed, Cursor, Windsurf — pick whatever, keep whatever, use cctk in the terminal alongside.
  5. Progressive disclosure. New users get a sensible starter set. Advanced teams can pull the whole 121-tool catalog. The learning curve stays gentle at both ends.

What's next

The roadmap for the next few months:

Templates (Q3): full CLAUDE.md templates for common stacks — Next.js, Django, Rails, Rust, Go. Not just commands, but the full "how our team works with Claude Code" convention set.

Guides (Q3): tutorial content on how to write your own commands, agents, and hooks. We want cctk to be the reference material even for teams that don't use our specific commands.

Team dashboard (Q4): a UI for teams on the paid plan to see command usage, share private commands across the org, and manage installations centrally.

Enterprise features (Q4): SSO, audit logs, private catalogs, on-premises hosting options for teams that need them.

How to get involved

cctk is community-driven. If you've built a slash command, subagent, or hook that works well for your team, contribute it. Every accepted contribution ships with attribution.

Newsletter: subscribe below for weekly command updates. Discord: discord.gg/cctk for real-time discussion.

Try it out: npx cctk@latest init in your favorite project. Five minutes to see if it fits. If it does, you've just saved yourself the "everyone builds the same commands" learning curve. If it doesn't, you've learned something about what your team needs that cctk isn't giving you — and we'd love to hear about it.

Welcome to cctk. Ship better with Claude Code.

MK

Written by

Mubashir K.

Mubashir writes about testing, refactoring, and the CLI workflows that keep Claude Code teams shipping.

Get cctk running in one command

85 slash commands, 12 subagents, 12 MCP integrations, 12 hooks. All the patterns from this post are shipped in cctk.

npx cctk@latest init Get cctk v1.5.2 →

Share with