Claude Code Multi-Agent Collaboration Setup

This repository contains workflows and scripts for enabling multiple AI agents to work together collaboratively using Claude Code's MCP (Model Context Protocol) feature.

Overview

With this setup, you can orchestrate multiple Claude Code instances that:

• Work on different aspects of a task in parallel
• Share discoveries through a chat room
• Review each other's work
• Automatically iterate until quality standards are met

Key Features

• Multi-Agent Bug Fix Workflow: Automated investigation → implementation → review → fix cycles
• Multi-Agent Investigation Workflow: Parallel investigation of complex systems with specialized agents
• Cost Optimization: 85%+ token cost reduction through prompt caching
• Quality Assurance: Automated code review with iterative improvements

Quick Start

1. Clone this repository

git clone https://github.com/mkXultra/claude_code_setup
cd claude_code_setup

2. Link to your project

In your working project directory:

ln -s /path/to/claude_code_setup guide

3. Install MCP components

From your project root:

./guide/mcp_add.sh

This installs:

• Chat MCP: For agent communication
• CCM (Claude Code Manager): For spawning and managing multiple Claude instances
• Playwright MCP: For browser automation (optional)

4. Verify installation

In Claude Code:

/mcp

You should see chat, ccm, and playwright listed.

5. Use a workflow

For bug fixing:

@guide/multi-agent-bug-fix-workflow.md Please fix [bug description]

For investigation:

@guide/multi-agent-investigation-workflow.md Please investigate [topic]

Workflows

Multi-Agent Bug Fix Workflow

Specialized agents working together:

1. Investigation Agent (Opus): Analyzes bug and proposes solutions
2. Implementation Agent (Sonnet): Implements the fix
3. Review Agent (Opus): Reviews code changes
4. Debug Agent (Sonnet): Troubleshoots when errors occur

The agents automatically cycle through review-fix iterations until code quality standards are met.

Multi-Agent Investigation Workflow

For complex system analysis:

• Divides investigation into 4-6 specialized areas
• Agents work in parallel on different aspects
• Dynamic agent addition based on discoveries
• Final integration by Opus model for high-quality reports

Example Results

Bug Fix Example

• Task: Fix file locking directory creation bug
• Time: 15 minutes (vs hours manually)
• Cost: 85% reduction through caching
• Quality: Automated review caught edge cases

Investigation Example

• Task: Permission system analysis
• Time: 20 minutes with 7 agents
• Deliverables: 5 comprehensive reports totaling 45KB
• Coverage: Complete system understanding

Cost Optimization

The workflows use a "foundation session" strategy:

{
  "cache_read_input_tokens": 390302,  // Reused tokens (cheap!)
  "input_tokens": 56                  // New tokens (expensive)
}

Result: 90%+ token cost savings

Files in this Repository

• mcp_add.sh: Installation script for MCP components
• multi-agent-bug-fix-workflow.md: Bug fixing workflow (English)
• multi-agent-bug-fix-workflow_jp.md: Bug fixing workflow (Japanese)
• multi-agent-investigation-workflow.md: Investigation workflow (English)
• multi-agent-investigation-workflow_jp.md: Investigation workflow (Japanese)
• multi-agent-feature-implementation-workflow.md: Feature implementation workflow

Requirements

• Claude Code with MCP support
• Node.js (for npx commands)
• Active Claude API access

Contributing

Feel free to submit issues and enhancement requests!

License

[Your License Here]

Acknowledgments

Built using:

• Claude Code MCP
• Agent Communication MCP
• Playwright MCP