MCP Mac Apps Server

# Ollama Integration in MCP Mac Apps Server ## ✅ What's Added Ollama API integration has been added to the MCP server, allowing you to use local LLM models directly from MCP tools. ### New Tools: 1. **`ollama_generate`** - Generate responses using local Ollama models 2. **`ollama_list_models`** - Get list of available Ollama models ## 🚀 Usage ### 1. Make Sure Ollama is Running ```bash # Check that Ollama server is running curl http://localhost:11434/api/tags # If not running, start it: ollama serve ``` ### 2. Rebuild MCP Server ```bash npm run build ``` ### 3. Usage via LLM Client After configuring in Claude Desktop or another MCP client, you can use the new tools: **Example Requests:** - "Use Ollama to explain this code" - "Show list of available Ollama models" - "Generate response using llama3.2 model" ## 📋 Tool Parameters ### `ollama_generate` **Parameters:** - `prompt` (required) - Prompt for the model - `model` (optional) - Model name (default: "llama3.2") **Usage Example:** ```json { "name": "ollama_generate", "arguments": { "prompt": "Explain what MCP protocol is", "model": "llama3.2" } } ``` ### `ollama_list_models` **Parameters:** none **Returns:** List of available models with sizes ## ⚙️ Configuration ### Changing Ollama Server URL By default, `http://localhost:11434` is used. To change URL, set environment variable: ```bash export OLLAMA_API_URL=http://your-ollama-server:11434 ``` Or change in code: ```typescript const OLLAMA_API_URL = "http://your-custom-url:11434"; ``` ## 🔧 Usage Examples ### Via Claude Desktop After configuring MCP server, simply ask Claude: ``` "Use Ollama to generate response to question: what is artificial intelligence?" ``` Claude will automatically use the `ollama_generate` tool. ### Combining with Other Tools You can combine Ollama with application management tools: ``` "Use Ollama to analyze contents of file ~/Documents/report.txt, then open TextEdit to show results" ``` ## 📊 Available Models Check list of models via `ollama_list_models` tool or manually: ```bash ollama list ``` Popular models: - `llama3.2` - fast, lightweight model (2GB) - `llama3.1:8b` - more powerful version (4.7GB) - `deepseek-r1:8b` - for reasoning (5.2GB) - `mistral:7b` - Mistral AI model - `qwen2.5:7b` - Alibaba Qwen ## 🛠️ Troubleshooting ### Error: "Failed to Connect to Ollama Server" **Solution:** 1. Make sure Ollama is running: `ollama serve` 2. Check that port 11434 is accessible: `curl http://localhost:11434/api/tags` 3. Check `OLLAMA_API_URL` environment variable ### Error: "Model Not Found" **Solution:** 1. Load model: `ollama pull llama3.2` 2. Check model list: `ollama list` ### Slow Generation **Causes:** - Model too large for your hardware - Insufficient RAM - CPU instead of GPU **Solutions:** - Use smaller model (e.g., `llama3.2` instead of `llama3.1:8b`) - Close other applications to free memory - On Mac with Apple Silicon, use Metal for acceleration ## 🔐 Security - Ollama API works locally and doesn't send data to internet - All requests are processed on your computer - Make sure Ollama server is not accessible externally (localhost only by default) ## 📚 Additional Information - [Ollama Documentation](https://ollama.ai/docs) - [Ollama API](https://github.com/ollama/ollama/blob/main/docs/api.md) - [Ollama Model List](https://ollama.ai/library)