ecc

Compose ClaudeCode

Multi-model agent orchestration for Claude Code with intelligent configuration and workflow management.

Quick Start

# 1. Install plugin
/plugin install compose-claudecode

# 2. (Optional) Setup MCP servers globally - ONE TIME for ALL projects
npx ecc mcp-setup

# 3. Initialize for your project
/setup

# 4. Start building
# Example: "Add user authentication"
# Example: "Create a dashboard page"

That's it! The /setup command will interview you about your project and automatically configure everything.

MCP Setup (npx ecc mcp-setup):

• Configure once, use in ALL projects
• Just paste your API keys
• Automatically updates ~/.claude.json
• No manual JSON editing needed!

---

Features

• 🎯 Smart Initialization: /setup command configures your project automatically
• 🤖 24 Specialized Agents: Optimal model assignments (40% token savings)
• ⚡ Two Modes: Auto (flexible) or Structured (predefined workflows)
• 📊 Multi-Model: Opus 4.5, Gemini 3, GLM-4.7, GPT-5.2
• 💰 Cost Optimized: Intelligent model selection saves tokens
• 🔧 Reconfigurable: Change settings anytime with /setup

---

Configuration

All configuration happens in two files:

1. CLAUDE.md (your project) - Generated by /setup

   • Your tech stack
   • Workflow preferences
   • Custom rules

2. config/stack.yaml (your project) - Auto-configured

   • Model assignments
   • Stack versions

Do not modify this plugin's README.md - it's just documentation. Do modify your project's CLAUDE.md - it's your configuration.

---

MCP Setup (One-Time, Optional)

Note: MCP is optional. This plugin works with Claude's built-in models (Opus, Sonnet, Haiku) without MCP configuration.

Automated Setup (Recommended) 🚀

One command, works for ALL projects:

npx ecc mcp-setup

This interactive command will:

1. Ask which MCP servers you want (Gemini, OpenAI, GLM)
2. Guide you to get API keys
3. Automatically update ~/.claude.json (global config)
4. Available in all projects immediately after restart

Benefits:

• ✅ Configure once, use everywhere
• ✅ No manual JSON editing
• ✅ Automatic backup before changes
• ✅ Secure (API keys hidden during input)

Manual Setup (Alternative)

If you prefer manual configuration:

1. Get API Keys First

• Gemini (Free): aistudio.google.com/apikey - 60 req/min, 1,500/day
• OpenAI (Paid): platform.openai.com/api-keys - GPT-4o, o1, o3
• GLM (Paid): open.bigmodel.cn - 30% cheaper

2. Choose Your Setup Method

Method A: Using Commands (Quick)

Run these commands and replace with your actual API keys:

# Add Gemini
claude mcp add gemini --command "npx" --args "-y" --args "github:aliargun/mcp-server-gemini" -e GEMINI_API_KEY="your-gemini-key-here"

# Add OpenAI (optional)
claude mcp add openai --command "npx" --args "-y" --args "@mzxrai/mcp-openai@latest" -e OPENAI_API_KEY="your-openai-key-here"

# Add GLM (optional)
claude mcp add glm --command "npx" --args "-y" --args "@zhipuai/mcp-server" -e ZHIPUAI_API_KEY="your-glm-key-here"

# Verify
claude mcp list

Method B: Manual Configuration

Edit ~/.claude.json and add the mcpServers section:

{
  "mcpServers": {
    "gemini": {
      "command": "npx",
      "args": ["-y", "github:aliargun/mcp-server-gemini"],
      "env": {
        "GEMINI_API_KEY": "your-gemini-key-here"
      }
    },
    "openai": {
      "command": "npx",
      "args": ["-y", "@mzxrai/mcp-openai@latest"],
      "env": {
        "OPENAI_API_KEY": "your-openai-key-here"
      }
    }
  }
}

Important: If ~/.claude.json already has content, merge the "mcpServers" section into your existing file.

Method C: Environment Variables (Most Secure)

1. Add to ~/.zshrc or ~/.bashrc:

   export GEMINI_API_KEY="your-gemini-key"
   export OPENAI_API_KEY="your-openai-key"
   export ZHIPUAI_API_KEY="your-glm-key"
   

2. Reload shell: source ~/.zshrc

3. Add MCP servers without hardcoding keys:

   claude mcp add gemini --command "npx" --args "-y" --args "github:aliargun/mcp-server-gemini"
   claude mcp add openai --command "npx" --args "-y" --args "@mzxrai/mcp-openai@latest"
   

3. Verify

claude mcp list
# Should show: gemini ✓ Connected (and others if added)

---

## How It Works

User Request ↓ Orchestrator analyzes ↓ ├─ Auto Mode: Creates plan on-the-fly └─ Structured Mode: Uses predefined workflow ↓ Agents execute (with optimal models) ↓ Checkpoints for review ↓ Done!

---

## Usage

### Automatic Execution

Everything runs automatically. Just make a request and the system handles it:

You: "Add user authentication"