OpenRouter MCP Server

# Custom Model Usage Guide This guide explains how to use custom OpenRouter models with the MCP server through natural language commands. ## Overview The OpenRouter MCP server now supports using any model available on OpenRouter through a custom model interface. This allows you to use specific models not covered by the standard aliases. ## Natural Language Commands You can use commands like: - "Claude, call the OpenRouter MCP and let me input the model myself" - "I want to use a custom model with OpenRouter" - "Let me specify my own OpenRouter model" - "Use a specific model code with OpenRouter" ## How It Works When you use these natural language commands, Claude Code will: 1. Launch the OpenRouter MCP server (if not already running) 2. Use the `chat_with_custom_model` tool 3. Ask you for the specific model code you want to use 4. Handle the conversation with your chosen model ## Supported Model Codes You can use any model available on OpenRouter. Examples include: - `openai/gpt-5` (latest flagship model with 400K context) - `deepseek/deepseek-chat-v3.1` (DeepSeek v3.1 with 163K context) - `anthropic/claude-3-opus` - `anthropic/claude-3-sonnet` - `meta-llama/llama-3.3-70b-instruct` - `mistralai/mixtral-8x7b-instruct` - `openai/gpt-4-turbo` - `google/gemini-pro` - And many more... ## Example Usage ### Basic Usage ``` You: "Claude, call the OpenRouter MCP and let me input the model myself" Claude: "I'll help you use a custom OpenRouter model. What model code would you like to use?" You: "meta-llama/llama-3.3-70b-instruct" Claude: [Uses the specified model for the conversation] ``` ### With Context ``` You: "I want to use openai/gpt-5 to analyze this large codebase" Claude: [Automatically uses gpt-5 model with 400K context for your files] ``` ### Using DeepSeek v3 ``` You: "Use deepseek-v3.1 for this complex reasoning task" Claude: [Automatically uses DeepSeek Chat v3.1 with 163K context] ``` ### Advanced Usage ``` You: "Use anthropic/claude-3-opus to analyze this code file" Claude: [Automatically uses claude-3-opus model with your file] ``` ## Features The custom model tool supports: - **Custom model selection**: Use any OpenRouter model by its exact code - **Conversation continuity**: Continue existing conversations with the same model - **File attachments**: Include code files for context - **Image support**: Attach images for models with vision capabilities - **Parameter control**: Optionally specify temperature and max_tokens - **Full context**: All conversation history is maintained ## Technical Details The `chat_with_custom_model` tool accepts: - `prompt` (required): Your message - `custom_model` (required): The exact OpenRouter model code - `continuation_id` (optional): To continue an existing conversation - `files` (optional): Array of file paths to include - `images` (optional): Array of image paths for vision models - `temperature` (optional): 0.0-2.0, default 0.7 - `max_tokens` (optional): Maximum response tokens ## Benefits 1. **Flexibility**: Use any model available on OpenRouter 2. **Experimentation**: Try different models for different tasks 3. **Specific needs**: Use specialized models for particular use cases 4. **Cost control**: Choose models based on your budget 5. **Latest models**: Access newly released models immediately ## Notes - Make sure you have a valid OpenRouter API key configured - Model availability and pricing varies - check OpenRouter's documentation - Some models may have specific capabilities (vision, function calling, etc.) - Response quality and speed varies by model