MCP AI Gateway

# How to Access OpenRouter's Free Gemini Models in Claude Desktop Save up to 80% on AI costs by accessing free and discounted models through OpenRouter while enjoying the familiar Claude Desktop interface. ## Problem Statement Claude Desktop is an excellent AI interface, but you're limited to Anthropic's models and pricing. Meanwhile, OpenRouter offers: - **Free tier models** including Gemini variants - **Significantly lower costs** than direct API access - **400+ models** from multiple providers - **Automatic fallbacks** when models are unavailable This guide shows you how to access OpenRouter's cost-effective models directly within Claude Desktop using MCP AI Gateway. ## Prerequisites - Claude Desktop installed - OpenRouter account (free signup at [openrouter.ai](https://openrouter.ai)) - Basic familiarity with JSON configuration ## Step 1: Get Your OpenRouter API Key 1. Sign up at [openrouter.ai](https://openrouter.ai) 2. Navigate to your API Keys section 3. Create a new API key (starts with `sk-or-`) 4. Note your free monthly credits (usually $5-10) ## Step 2: Configure MCP AI Gateway for OpenRouter Add this configuration to your Claude Desktop MCP settings: ```json { "mcpServers": { "openrouter-free": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "openai", "API_KEY": "sk-or-your-openrouter-key-here", "API_ENDPOINT": "https://openrouter.ai/api/v1", "DEFAULT_MODEL": "google/gemini-flash-1.5", "DESCRIPTION": "Free & Low-Cost Models via OpenRouter:\n\n💰 FREE TIER:\n- google/gemini-flash-1.5: Fast responses, completely free\n- microsoft/phi-3-mini-4k-instruct: Small but capable\n- meta-llama/llama-3.1-8b-instruct: Open source powerhouse\n\n💡 ULTRA-LOW COST:\n- google/gemini-pro-1.5: $0.00125/1K tokens (vs $0.015 direct)\n- anthropic/claude-3-haiku: $0.00025/1K tokens\n- openai/gpt-3.5-turbo: $0.0005/1K tokens\n\n🚀 PREMIUM DISCOUNTED:\n- anthropic/claude-3.5-sonnet: 20% cheaper than direct\n- openai/gpt-4o: Competitive pricing with better availability" } } } } ``` ## Step 3: Configure Your Claude Desktop Settings ### On macOS: ```bash ~/Library/Application Support/Claude/claude_desktop_config.json ``` ### On Windows: ```bash %APPDATA%\Claude\claude_desktop_config.json ``` Edit the file and add the OpenRouter configuration above. ## Step 4: Restart Claude Desktop After saving the configuration: 1. Completely quit Claude Desktop 2. Restart the application 3. You should see the new `chat_completion` tool available ## Step 5: Test Your Setup Try this prompt in Claude Desktop: > "Please use the OpenRouter gateway to compare the free Gemini Flash model with a premium model for writing a product description for wireless headphones." Claude will automatically use the OpenRouter models based on your DESCRIPTION guidance. ## Cost Comparison Analysis | Model | Direct API | OpenRouter | Savings | |-------|------------|------------|---------| | Gemini Flash 1.5 | $0.002/1K | **FREE** | 100% | | Gemini Pro 1.5 | $0.015/1K | $0.00125/1K | 92% | | Claude 3 Haiku | $0.00025/1K | $0.00025/1K | Same | | GPT-4o | $0.005/1K | $0.004/1K | 20% | **Monthly Usage Example:** - 1M tokens/month through OpenRouter: ~$1.25 - Same usage direct APIs: ~$15 - **Total savings: $13.75/month (92%)** ## Advanced Configuration: Smart Cost Optimization Set up multiple endpoints for maximum cost efficiency: ```json { "mcpServers": { "free-tier": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "openai", "API_KEY": "sk-or-your-key", "API_ENDPOINT": "https://openrouter.ai/api/v1", "DEFAULT_MODEL": "google/gemini-flash-1.5", "DESCRIPTION": "FREE MODELS ONLY - Use for: simple queries, drafts, brainstorming" } }, "premium-discounted": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "openai", "API_KEY": "sk-or-your-key", "API_ENDPOINT": "https://openrouter.ai/api/v1", "DEFAULT_MODEL": "anthropic/claude-3.5-sonnet", "DESCRIPTION": "PREMIUM MODELS - Use for: complex analysis, code review, important content" } } } } ``` ## Troubleshooting ### Common Issues **1. "Model not found" error:** ```json // Check model names at openrouter.ai/models "DEFAULT_MODEL": "google/gemini-flash-1.5" // Correct format ``` **2. Authentication errors:** ```bash # Verify your API key format sk-or-v1-1234... ✓ Correct sk-1234... ✗ Wrong (OpenAI key) ``` **3. Tool not appearing:** - Ensure Claude Desktop is completely restarted - Check JSON syntax in configuration file - Verify file permissions ### Rate Limiting OpenRouter free tier limits: - **Requests**: 200/minute - **Tokens**: 10,000/day - **Monthly**: $10 worth of credits Monitor usage at: openrouter.ai/usage ## Performance Optimization Tips ### 1. Model Selection Strategy ``` Simple tasks → Gemini Flash (free) Medium complexity → Gemini Pro 1.5 (ultra-low cost) Complex analysis → Claude 3.5 Sonnet (discounted premium) ``` ### 2. Prompt Engineering for Cost - Be specific in model requests: "Use the free Gemini model for this simple task" - Batch similar requests together - Use shorter prompts when possible ### 3. Monitoring and Budgeting - Set OpenRouter spending limits - Review usage patterns monthly - Identify which tasks can use free models ## Real-World Usage Examples ### Example 1: Content Creation Workflow ``` 1. Brainstorm ideas → Gemini Flash (free) 2. Write first draft → Gemini Pro 1.5 (low-cost) 3. Final edit → Claude 3.5 Sonnet (premium) ``` ### Example 2: Development Tasks ``` 1. Code suggestions → Gemini Flash (free) 2. Code review → Claude 3.5 Sonnet (premium) 3. Documentation → Gemini Pro 1.5 (low-cost) ``` ## Next Steps 1. **Monitor Your Usage**: Check OpenRouter dashboard weekly 2. **Experiment with Models**: Try different models for various tasks 3. **Scale Gradually**: Increase usage as you find your optimal model mix 4. **Set Up Alerts**: Configure spending alerts in OpenRouter ## Advanced: Multiple Provider Setup For maximum flexibility, combine OpenRouter with direct APIs: ```json { "mcpServers": { "openrouter-budget": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "openai", "API_KEY": "sk-or-your-key", "API_ENDPOINT": "https://openrouter.ai/api/v1", "DESCRIPTION": "Budget-friendly options via OpenRouter" } }, "anthropic-direct": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "anthropic", "API_KEY": "sk-ant-your-key", "DESCRIPTION": "Direct Anthropic access for guaranteed performance" } } } } ``` This setup gives you the best of both worlds: cost optimization through OpenRouter and reliable access through direct APIs. ## Conclusion By integrating OpenRouter with Claude Desktop, you can: - **Reduce costs by up to 92%** while keeping the same interface - **Access 400+ models** from a single familiar interface - **Maintain quality** with smart model selection - **Scale efficiently** as your usage grows Start with the free tier, monitor your usage patterns, and gradually optimize your model selection strategy for maximum cost efficiency.