Total: 4,734 lines · 157 KB of curated, production-verified reference material

🗺️ How to Navigate This Repo

New to Claude Code?

commands.md → mcp-servers.md → plugins.md

Learn what Claude can do natively → connect it to the world → extend with skills.

Building a production app?

tools.md → workflows.md → agent-frameworks.md

Pick your stack → follow the workflow pattern → choose the right orchestration.

Looking for something specific?

| I want to... | Go to | |-------------|-------| | Learn all Claude commands | commands.md | | Connect Claude to GitHub / Postgres / Slack | mcp-servers.md | | Give Claude persistent memory | mcp-servers.md#memory-systems | | Install the best plugins | plugins.md | | Build a RAG pipeline | workflows.md | | Compare CrewAI vs LangGraph | agent-frameworks.md | | Pick a vector database | tools.md | | Deploy Claude to production | tools.md + workflows.md |

Claude OS — Layer Model

This is a framework, not just a collection.

1. Commands — Interaction
2. MCP — Execution
3. Plugins — Capabilities
4. Tools — Infrastructure
5. Workflows — Automation
6. Agents — Intelligence scaling

Start Here If You Have 10 Minutes

• Install 3 plugins
• Add Memory MCP
• Try Builder–Validator

⚡ Quick Wins — 5 Things to Do Right Now

1. Install the 3 plugins that matter most

/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace      # Plans before it builds
/plugin install context7@claude-plugins-official         # Live docs, no hallucinated APIs
claude skills add juliusbrussee/caveman                  # 65-75% fewer output tokens

2. Set up your core MCP servers

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/projects"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "your_token" }
    },
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"]
    }
  }
}

3. Initialize project memory

/init           # Creates CLAUDE.md — Claude reads this every session
/memory         # Edit it with your stack, conventions, and rules

4. Learn the 5 keyboard shortcuts that save the most time

Shift+Tab    → Cycle modes: Normal → Auto-Accept → Plan Mode
Ctrl+R       → Search command history
Esc Esc      → Rewind conversation or code
Ctrl+T       → Toggle task list
/compact     → Compress context when it hits 70%

5. Use the right thinking mode for the right task

L99          → Maximum reasoning depth (architecture decisions)
XRAY         → Root cause analysis (debugging)
OODA         → Tactical decisions under uncertainty
BEASTMODE    → Maximum effort, no shortcuts
/plan        → Force planning before execution (production code)

📊 The Ecosystem at a Glance

MCP Servers — By Category

| Category | Must-Have Servers | |---------|------------------| | Version Control | GitHub, Git, GitLab | | Database | PostgreSQL, SQLite, Redis, Supabase, MongoDB | | Communication | Slack, Notion, Gmail, Discord | | Design | Figma | | Deployment | Vercel, Docker, AWS, Kubernetes | | Search | Brave Search, Fetch, Puppeteer | | Memory | Memory MCP, mem0, Memora, OmniMem |

Plugins — By Role

| Role | Essential Installs | |-----|-------------------| | All developers | Superpowers, Context7, Caveman | | Full-stack | + feature-dev, frontend-design, Playwright | | Backend | + backend-architect, test-writer-fixer, security-sweep | | Product managers | + Taskmaster, backlog, BMAD Method | | Security | + security-sweep, AgentLint, Trail of Bits | | Content/Marketing | + connect-apps, marketing-skills |

Agent Frameworks — Decision Matrix

| Need | Framework | |------|----------| | Fastest to demo | CrewAI | | Production + state management | LangGraph | | Research + code execution | AutoGen | | Claude-only stack | Claude Agent SDK | | Data-heavy agents | LlamaIndex Agents | | Enterprise .NET/Azure | Microsoft Agent Framework |

Tools — Stack by Use Case

| Use Case | Stack | |---------|-------| | Getting started | Chroma + LangChain + Streamlit | | Production SaaS | Supabase + FastAPI + Pinecone + Vercel | | Enterprise | LlamaIndex + LangGraph + Qdrant + Kubernetes | | No-code | Flowise or Dify + Zapier or Make | | Privacy-first | Ollama + Chroma + LlamaIndex (fully local) |

🏗️ Repo Structure

claude-os/
│
├── README.md                  ← You are here
│
├── commands.md                ← Slash commands, CLI flags, shortcuts, prompt codes
├── mcp-servers.md             ← MCP servers (Tier 1–3) + memory systems
├── plugins.md                 ← Plugins, skills, extensions (Tier 1–6)
├── tools.md                   ← Frameworks, DBs, deployment, automation
├── workflows.md               ← 25 production workflows with steps + tools
└── agent-frameworks.md        ← 25 frameworks with benchmarks + code examples

🔢 Numbers That Matter

| Metric | Number | |--------|--------| | Commands documented | 100+ | | MCP servers covered | 25 (from 1,200+ ecosystem) | | Memory systems | 10 | | Plugins and skills | 40+ | | Tools documented | 25 | | Workflows with full steps | 25 | | Agent frameworks | 25 | | Code examples included | 60+ | | Total lines of reference | 4,700+ | | Repos indexed as sources | 14,634 |

🧭 The Philosophy Behind This Bundle

Most Claude resources are either:

Too shallow — "Here are 10 cool things Claude can do" (no install commands, no real usage)

Too scattered — spread across 50 different GitHub repos, Medium articles, and Discord threads

Too outdated — written for Claude 2 or early MCP, missing everything from 2025–2026

Claude OS is different. Every entry follows the same structure:

• What it does (one clear sentence)
• Why it matters for Claude specifically
• Working install command or code example
• Honest "best for" and limitations
• Updated to April 2026

The rule we followed throughout:

If you can't explain what a tool does in one sentence and show working code in 10 lines, it doesn't belong here.

📈 Verified Production Metrics

Real companies. Real results. What's possible when you use this stack properly.

| Company | Workflow | Result | |---------|---------|--------| | Stripe | Claude Code across 1,370 engineers | 10k-line migration: 4 days vs 10 engineer-weeks | | Ramp | Incident investigation workflow | 80% reduction in investigation time | | Wiz | Multi-agent code migration | 50k lines Python→Go in ~20hrs (was 2–3 months) | | Rakuten | Feature delivery pipeline | 24 working days → 5 working days | | Fountain | Multi-agent code review | 50% faster delivery | | CRED | Agent teams across dev lifecycle | 2x speed, maintained quality |

🗓️ Keeping This Current

Claude Code ships weekly. The ecosystem moves fast.

How to know if something is outdated:

• Run /help in Claude Code — always shows current command list
• Check claude --version before using CLI flags
• MCP server transport syntax changed in 2026 (use --transport http for cloud servers)

Last full audit: April 28, 2026 Next planned update: When Claude Code hits v3.0 or major ecosystem shift

🤝 Contributing

Found a command that works but isn't listed? Know a plugin that deserves Tier 1? Discovered a workflow pattern that ships faster?

Open a PR with:

• Name and one-line description
• Working install command or code example
• "Best for" and honest l