MCP Mac Apps Server

# Ollama Setup with MCP Mac Apps Server ## ✅ Current Status - ✅ Ollama installed - ✅ Ollama server running - ✅ Model `llama3.2` loaded (2.0 GB) - ✅ Model `deepseek-r1:8b` available (5.2 GB) ## 🚀 Quick Start ### Option 1: Usage via Claude Desktop 1. **Install Claude Desktop** (if not already installed): - Download from https://claude.ai/download 2. **Install MCP Server for Ollama**: ```bash npx -y @modelcontextprotocol/create-server ollama-mcp ``` Or add to Claude Desktop configuration (`~/Library/Application Support/Claude/claude_desktop_config.json`): ```json { "mcpServers": { "ollama": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-ollama"] }, "mac-apps": { "command": "node", "args": ["/Users/olegzaichkin/Documents/MCP/dist/index.js"] } } } ``` 3. **Restart Claude Desktop** 4. **Now Claude can**: - Use local models via Ollama - Manage Mac applications through your MCP server ### Option 2: Direct API Usage Ollama provides REST API at `http://localhost:11434`. You can use it directly with any client that supports OpenAI-compatible API. **Testing API:** ```bash curl http://localhost:11434/api/generate -d '{ "model": "llama3.2", "prompt": "Hello! How are you?", "stream": false }' ``` ## 📝 Available Models Check model list: ```bash ollama list ``` Load other models: ```bash # Popular models ollama pull llama3.1:8b # More powerful version ollama pull mistral:7b # Mistral AI ollama pull qwen2.5:7b # Alibaba Qwen ollama pull codellama:7b # For programming ollama pull phi3 # Lightweight Microsoft model ``` ## 🔧 Ollama Management **Start Server:** ```bash ollama serve ``` **Stop Server:** ```bash # Press Ctrl+C or find process and kill it ps aux | grep ollama kill <PID> ``` **Auto-start (macOS):** Ollama usually starts automatically via LaunchAgent. If you need to add to autostart: ```bash # Create LaunchAgent cat > ~/Library/LaunchAgents/com.ollama.server.plist << EOF <?xml version="1.0" encoding="UTF-8"?> <!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd"> <plist version="1.0"> <dict> <key>Label</key> <string>com.ollama.server</string> <key>ProgramArguments</key> <array> <string>/opt/homebrew/bin/ollama</string> <string>serve</string> </array> <key>RunAtLoad</key> <true/> <key>KeepAlive</key> <true/> </dict> </plist> EOF # Load agent launchctl load ~/Library/LaunchAgents/com.ollama.server.plist ``` ## 🎯 Usage Examples ### Test Model Directly: ```bash ollama run llama3.2 "Tell me about MCP protocol" ``` ### Usage via API with curl: ```bash # Simple request curl http://localhost:11434/api/generate -d '{ "model": "llama3.2", "prompt": "What is Model Context Protocol?", "stream": false }' # With streaming curl http://localhost:11434/api/generate -d '{ "model": "llama3.2", "prompt": "Hello!", "stream": true }' ``` ### Usage with Python: ```python import requests import json response = requests.post( 'http://localhost:11434/api/generate', json={ 'model': 'llama3.2', 'prompt': 'Open Safari', 'stream': False } ) print(response.json()['response']) ``` ## 🔍 Verification Check that server is running: ```bash curl http://localhost:11434/api/tags ``` Should return list of models in JSON format. ## 💡 Tips 1. **Performance**: Model `llama3.2` (2GB) works fast but is less powerful. For better quality, use `llama3.1:8b` or `deepseek-r1:8b`. 2. **Memory**: Make sure you have enough RAM. Models require: - `llama3.2`: ~2-4 GB RAM - `llama3.1:8b`: ~8-10 GB RAM - `deepseek-r1:8b`: ~10-12 GB RAM 3. **Speed**: On Mac with Apple Silicon (M1/M2/M3), models work significantly faster thanks to the neural processor. 4. **Privacy**: All processing happens locally, data is not sent anywhere. ## 📚 Useful Links - [Ollama Documentation](https://ollama.ai/docs) - [Available Models](https://ollama.ai/library) - [Ollama GitHub](https://github.com/ollama/ollama)