Multi-Agent Coding Environment (Gemini, Claude, Codex)

This repository provides the necessary components to set up a powerful multi-agent coding environment. In this setup, Gemini CLI acts as the orchestrator, delegating complex coding and planning tasks to specialized AI models like Claude Code and OpenAI Codex via custom Model Context Protocol (MCP) bridge servers.

🎯 Overview

This project addresses the challenge of leveraging the unique strengths of different AI models in a collaborative development workflow. Gemini, with its orchestration capabilities, can intelligently route specific tasks to the most suitable AI agent:

• Claude Code: Ideal for brainstorming, architectural design, code analysis (security, performance, maintainability), and collaborative planning.
• OpenAI Codex: Excellent for code generation, scaffolding new features, refactoring, and executing code within a sandboxed environment.

This creates a highly efficient and versatile AI-assisted development pipeline.

🏗️ Architecture

User Query
    ↓
Gemini CLI (Orchestrator)
    ↓
MCP Bridge Servers
    ├───► Claude Code Bridge (claude-mcp) ───► Claude Code
    └───► OpenAI Codex Bridge (codex-mcp) ───► OpenAI Codex CLI

✨ Features

By integrating these agents, you can:

• Brainstorm and Plan: Utilize Claude Code for high-level architectural discussions, design pattern considerations, and detailed implementation planning.
• Code Generation & Refactoring: Delegate code scaffolding, feature implementation, and refactoring tasks to OpenAI Codex.
• Code Analysis: Get in-depth security, performance, and maintainability analysis from Claude Code.
• Seamless Collaboration: Experience a unified workflow where Gemini intelligently manages interactions with specialized AI tools.

🚀 Getting Started

To set up this multi-agent environment, follow these steps:

1. Clone this repository:

   git clone <your-repo-url>
   cd mcp

2. Set up Claude Code Bridge:

   Navigate to the claude-mcp directory and follow the instructions in its README.md for installation and configuration.

   cd claude-mcp
   # Follow instructions in claude-mcp/README.md

3. Set up OpenAI Codex Bridge:

   Navigate to the codex-mcp directory and follow the instructions in its README.md for installation and configuration.

   cd ../codex-mcp
   # Follow instructions in codex-mcp/README.md

4. Configure Gemini CLI:

   Ensure your Gemini CLI is configured to use both claude-bridge and codex-bridge as MCP servers. The setup scripts within each bridge's directory should assist with this configuration.

   Additionally, this repository includes GEMINI_COLLAB.md, an example of how to configure the Gemini CLI to adopt a new persona, specifically an "AI Project Coordinator" who orchestrates a team of AI engineers. To enable this collaborative mode, add GEMINI_COLLAB.md to your Gemini CLI configuration's contextFileName array, as shown below:

   {
     "theme": "Atom One",
     "selectedAuthType": "oauth-personal",
     "SystemPrompts": {
       "Software Engineer": "You are an experienced, world class, software engineer."
     },
     "contextFileName": [
       "GEMINI.md",
       "GEMINI_COLLAB.md",
       "GEMINI_EXAMPLES.md"
     ],
     "mcpServers": {
       "claude-bridge": {
         "command": "/home/ash/bin/claude-bridge",
         "args": [],
         "env": {},
         "timeout": 280000
       },
       "codex-bridge": {
         "command": "/home/ash/bin/codex-bridge",
         "args": [],
         "env": {},
         "timeout": 280000
       }
     },
     "preferredEditor": "vim"
   }

📂 Project Structure

• claude-mcp/: Contains the MCP bridge server for integrating Claude Code with Gemini CLI.
• codex-mcp/: Contains the MCP bridge server for integrating OpenAI Codex CLI with Gemini CLI.

🤝 Contributing

Contributions are welcome! Please refer to the individual README.md files within claude-mcp/ and codex-mcp/ for specific contribution guidelines related to each bridge server.

📄 License

This project is licensed under the MIT License. See the LICENSE file for details. (Note: A LICENSE file should be added to the root of this repository if not already present.)