MCP AI Gateway

# How to Use ChatGPT Models in Gemini CLI: A Complete Integration Guide Break free from Google's model limitations and access OpenAI's latest models directly within your familiar Gemini CLI workflow. ## Problem Statement Gemini CLI is a powerful development tool, but you're restricted to Google's Gemini models. What if you want to: - Use **GPT-4o's multimodal capabilities** for code analysis with screenshots - Access **GPT-5's superior reasoning** for complex architectural decisions - Compare **OpenAI vs Google models** side-by-side in the same interface - Leverage **OpenRouter's 400+ models** through Gemini CLI This guide shows you how to integrate any OpenAI-compatible API with Gemini CLI using MCP AI Gateway. ## Prerequisites - Gemini CLI installed and configured - OpenAI API key (or OpenRouter account) - Basic familiarity with MCP server configuration - Node.js 18+ for MCP AI Gateway ## Understanding Gemini CLI MCP Integration Gemini CLI uses a **reason and act (ReAct) loop** with built-in tools and MCP servers to complete complex tasks. The `/mcp` command lists configured servers and their available tools. Key capabilities: - **Rich content support**: Text, images, audio, binary data - **External system integration**: APIs, databases, file systems - **Multi-agent orchestration**: Complex workflows with tool chaining ## Step 1: Install and Configure MCP AI Gateway ### Quick Installation ```bash # Install globally for easy access npm install -g mcp-ai-gateway # Or use npx (recommended) npx mcp-ai-gateway --version ``` ### Create MCP Server Configuration Create a configuration file for Gemini CLI MCP integration: ```bash # Create MCP config directory mkdir -p ~/.config/gemini-cli/mcp ``` ### Configure for OpenAI Access Create `~/.config/gemini-cli/mcp/openai-gateway.json`: ```json { "mcpServers": { "openai-models": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "openai", "API_KEY": "sk-your-openai-key-here", "API_ENDPOINT": "https://api.openai.com/v1", "DEFAULT_MODEL": "gpt-4o", "DEFAULT_TEMPERATURE": "0.7", "DESCRIPTION": "OpenAI Models via Gemini CLI:\n\n🧠 REASONING & CODE:\n- gpt-4o: Best multimodal model with vision, audio, and code\n- gpt-4-turbo: Fast and capable for most development tasks\n- gpt-3.5-turbo: Quick responses for simple queries\n\n🎯 USE CASES:\n- Code review with screenshot analysis\n- Architecture design and planning\n- Cross-platform development insights\n- Alternative perspectives to Gemini models\n\n💡 INTEGRATION BENEFITS:\n- Compare OpenAI vs Gemini approaches\n- Leverage different model strengths\n- Redundancy when one service has issues" } } } } ``` ## Step 2: Configure Gemini CLI MCP Settings ### Update Gemini CLI Configuration Edit your Gemini CLI settings to include the MCP server: ```bash # Open Gemini CLI config gemini config edit ``` Add MCP server configuration: ```json { "mcp": { "servers": [ { "name": "openai-gateway", "command": ["npx", "mcp-ai-gateway"], "env": { "API_FORMAT": "openai", "API_KEY": "sk-your-openai-key-here", "API_ENDPOINT": "https://api.openai.com/v1", "DEFAULT_MODEL": "gpt-4o", "DESCRIPTION": "Access to OpenAI GPT models for comparison and specialized tasks" } } ] } } ``` ## Step 3: Verify Integration ### Check MCP Server Status ```bash # List configured MCP servers gemini /mcp # Expected output: # ✓ openai-gateway - Connected # Tools: chat_completion # Status: Active # Models: gpt-4o, gpt-4-turbo, gpt-3.5-turbo ``` ### Test Basic Functionality ```bash gemini chat "Please use the OpenAI gateway to explain the difference between async/await and Promises, then use Gemini to provide a Google Cloud specific example" ``` ## Step 4: Advanced Multi-Model Workflows ### Comparative Analysis Workflow Create sophisticated workflows that leverage both Gemini and OpenAI models: ```bash gemini chat " I need to design a microservices architecture. Please: 1. Use the OpenAI gateway (GPT-4o) to provide general microservices best practices 2. Use Gemini to suggest Google Cloud specific services and patterns 3. Compare both approaches and recommend a hybrid solution " ``` ### Code Review with Multiple Perspectives ```bash gemini review my-code.py --use-mcp openai-gateway --compare-with gemini ``` ### Cross-Model Validation ```bash gemini chat " For this React component, please: 1. Use GPT-4o to identify potential performance issues 2. Use Gemini to suggest Google-specific optimizations (like Core Web Vitals) 3. Synthesize both recommendations into actionable improvements " ``` ## Step 5: OpenRouter Integration for 400+ Models ### Configure OpenRouter Access ```json { "mcpServers": { "openrouter-models": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "openai", "API_KEY": "sk-or-your-openrouter-key", "API_ENDPOINT": "https://openrouter.ai/api/v1", "DEFAULT_MODEL": "anthropic/claude-3.5-sonnet", "DESCRIPTION": "400+ Models via OpenRouter:\n\n🏢 ENTERPRISE:\n- anthropic/claude-3.5-sonnet: Best coding model\n- openai/gpt-4o: Multimodal capabilities\n- google/gemini-pro-1.5: Advanced reasoning\n\n🆓 FREE TIER:\n- google/gemini-flash-1.5: Fast free responses\n- meta-llama/llama-3.1-8b: Open source\n- microsoft/phi-3-mini: Efficient small model\n\n💰 COST-EFFECTIVE:\n- Automatic fallbacks\n- Bulk pricing discounts\n- Free tier access" } } } } ``` ## Real-World Usage Examples ### Example 1: Full-Stack Development ```bash # Multi-model development workflow gemini chat " I'm building a Next.js app with Google Cloud backend. Please: 1. Use Claude 3.5 Sonnet (via OpenRouter) to review my React components for best practices 2. Use GPT-4o to suggest TypeScript improvements and type safety 3. Use Gemini to recommend Google Cloud services and deployment strategies 4. Synthesize all recommendations into a development roadmap " ``` ### Example 2: Technical Writing ```bash # Documentation with multiple perspectives gemini docs generate --topic "API Design" --use-models "gpt-4o,gemini-pro,claude-sonnet" ``` ### Example 3: Code Optimization ```bash gemini optimize performance.js --analyze-with openai-gateway --implement-with gemini --validate-with claude ``` ## Configuration Templates ### Template 1: Development Team Setup ```json { "mcpServers": { "ai-dev-tools": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "openai", "API_KEY": "sk-or-team-key", "API_ENDPOINT": "https://openrouter.ai/api/v1", "DEFAULT_MODEL": "anthropic/claude-3.5-sonnet", "DESCRIPTION": "Team Development Models:\n- Code review: Claude 3.5 Sonnet\n- Architecture: GPT-4o\n- Quick queries: Gemini Flash (free)\n- Documentation: Gemini Pro" } } } } ``` ### Template 2: Cost-Optimized Setup ```json { "mcpServers": { "budget-ai": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "openai", "API_KEY": "sk-or-budget-key", "API_ENDPOINT": "https://openrouter.ai/api/v1", "DEFAULT_MODEL": "google/gemini-flash-1.5", "DESCRIPTION": "Budget-Friendly AI:\n- Free tier: Gemini Flash, Llama 3.1\n- Low-cost: Claude Haiku, GPT-3.5\n- Premium only when needed: GPT-4o, Claude Sonnet" } } } } ``` ## Troubleshooting Common Issues ### 1. MCP Server Not Connecting ```bash # Check server status gemini /mcp status # Debug connection npx mcp-ai-gateway --test-connection # Verify environment variables echo $API_KEY | head -c 10 ``` ### 2. Model Not Found Errors ```bash # List available models curl -H "Authorization: Bearer $API_KEY" https://openrouter.ai/api/v1/models # Update configuration with correct model names "DEFAULT_MODEL": "openai/gpt-4o" # Correct OpenRouter format ``` ### 3. Authentication Issues ```bash # Test API key directly curl -H "Authorization: Bearer sk-..." https://api.openai.com/v1/models # Check key format OpenAI: sk-proj-... or sk-... OpenRouter: sk-or-v1-... Anthropic: sk-ant-... ``` ### 4. Performance Issues ```bash # Enable debug logging export DEBUG=mcp-ai-gateway:* gemini chat "test message" # Monitor token usage gemini usage --provider openai-gateway ``` ## Advanced Features ### Custom Model Routing Create intelligent model selection based on task type: ```json { "DESCRIPTION": "Smart Model Routing:\n\n📝 TEXT TASKS → GPT-4o\n🖼️ VISION TASKS → GPT-4o or Claude 3.5 Sonnet\n💻 CODE TASKS → Claude 3.5 Sonnet\n⚡ QUICK TASKS → Gemini Flash (free)\n🧮 MATH/LOGIC → GPT-4o or Gemini Pro\n🌐 WEB/API → Gemini Pro (Google expertise)\n\nAI will automatically choose the best model for each task type." } ``` ### Multi-Provider Fallbacks ```json { "mcpServers": { "primary-ai": { "env": { "API_ENDPOINT": "https://api.openai.com/v1", "DESCRIPTION": "Primary: OpenAI GPT models" } }, "fallback-ai": { "env": { "API_ENDPOINT": "https://openrouter.ai/api/v1", "DESCRIPTION": "Backup: 400+ models via OpenRouter" } } } } ``` ## Performance Optimization ### 1. Caching Strategy ```bash # Enable response caching export MCP_CACHE_ENABLED=true export MCP_CACHE_TTL=3600 # 1 hour ``` ### 2. Concurrent Requests ```bash # Allow parallel model calls export MCP_MAX_CONCURRENT=5 ``` ### 3. Model-Specific Timeouts ```json { "env": { "DEFAULT_TIMEOUT": "60000", # 60 seconds "FAST_MODEL_TIMEOUT": "15000" # 15 seconds for quick models } } ``` ## Conclusion By integrating OpenAI models with Gemini CLI through MCP AI Gateway, you get: - **Best of both worlds**: Google Cloud expertise + OpenAI capabilities - **Cost optimization**: Smart model selection and OpenRouter discounts - **Workflow efficiency**: Seamless model switching within familiar tools - **Flexibility**: 400+ models available through unified interface This setup transforms Gemini CLI from a single-model tool into a universal AI development platform, giving you the flexibility to choose the right model for each specific task while maintaining your existing workflow and expertise.