MCP Mac Apps Server

# Compatible LLMs and Clients for MCP Mac Apps Server Your MCP server works with **any LLM** that uses an MCP-compatible client. MCP (Model Context Protocol) is a standardized protocol, so the main thing is to use a client that supports it. ## 🔌 MCP-Compatible Clients ### 1. Claude Desktop (Anthropic) ⭐ Recommended **Status**: Full MCP support, official client **Installation**: - Download from the official Anthropic website - Install and create an account **Configuration**: 1. Find the file: `~/Library/Application Support/Claude/claude_desktop_config.json` 2. Add configuration: ```json { "mcpServers": { "mac-apps": { "command": "node", "args": ["/Users/olegzaichkin/Documents/MCP/dist/index.js"] } } } ``` **Available LLMs**: Claude 3.5 Sonnet, Claude 3 Opus, Claude 3 Haiku --- ### 2. Ollama (Local Models) 🏠 ✅ INSTALLED **Status**: Local models via MCP clients **Current models**: - ✅ `llama3.2` (2.0 GB) - installed - ✅ `deepseek-r1:8b` (5.2 GB) - installed 📖 **Detailed instructions**: [OLLAMA_SETUP.md](./OLLAMA_SETUP.md) **Installation**: ```bash # Install Ollama brew install ollama # Or download from https://ollama.ai # Start Ollama ollama serve ``` **Available models** (loaded on demand): ```bash # Popular models: ollama pull llama3.2 # 3B parameters, very fast ollama pull llama3.1:8b # 8B parameters ollama pull qwen2.5:7b # Alibaba Qwen ollama pull mistral:7b # Mistral AI ollama pull codellama:7b # For programming ollama pull phi3 # Microsoft Phi-3 ``` **Usage with MCP**: - Via Claude Desktop: install MCP server for Ollama - Or use other MCP clients with Ollama support --- ### 3. LM Studio (Local Models) 🎨 **Status**: Graphical interface for local LLMs **Installation**: - Download from https://lmstudio.ai - Install and load models through the interface **MCP Support**: - Official MCP server for LM Studio available - Integrates with Claude Desktop **Available models**: Supports GGUF format (Llama, Mistral, Phi, etc.) --- ### 4. DeepSeek **Status**: Depends on MCP support in DeepSeek client **Options**: - If DeepSeek Desktop supports MCP (see DEEPSEEK_SETUP.md) - Via API through other clients --- ### 5. OpenAI (GPT-4, GPT-3.5) **Status**: Via Claude Desktop or other MCP clients **Usage**: - Claude Desktop can use OpenAI API (if supported) - Or through other MCP-compatible clients --- ### 6. Other Local LLMs Through Ollama or LM Studio you can use: - **Llama 3** (Meta) - various sizes - **Mistral** (Mistral AI) - **Qwen** (Alibaba) - **Phi-3** (Microsoft) - **Code Llama** (Meta) - for code - **StarCoder** - for programming - And many others... --- ## 🎯 Selection Recommendations ### For Quick Start **Claude Desktop** - the simplest option: - ✅ Installation in 2 minutes - ✅ Full MCP support out of the box - ✅ Excellent response quality - ❌ Requires internet (cloud model) ### For Privacy and Offline Work **Ollama + Claude Desktop**: - ✅ Works completely locally - ✅ No internet required - ✅ Free - ❌ Requires sufficient RAM (8GB+ for large models) - ❌ May be slower than cloud models ### For Convenient Model Management **LM Studio**: - ✅ Beautiful graphical interface - ✅ Easy to switch between models - ✅ Detailed model information - ✅ MCP support through integration --- ## 🔧 Local Model Configuration ### Option 1: Ollama via MCP 1. Install Ollama and start the server 2. Install MCP server for Ollama in Claude Desktop: ```json { "mcpServers": { "ollama": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-ollama"] }, "mac-apps": { "command": "node", "args": ["/Users/olegzaichkin/Documents/MCP/dist/index.js"] } } } ``` ### Option 2: LM Studio MCP 1. Install LM Studio 2. Start local server in LM Studio 3. Add to Claude Desktop configuration: ```json { "mcpServers": { "lmstudio": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-lmstudio"] }, "mac-apps": { "command": "node", "args": ["/Users/olegzaichkin/Documents/MCP/dist/index.js"] } } } ``` --- ## 📊 Option Comparison | Option | Speed | Privacy | Quality | Cost | Internet | |--------|-------|---------|---------|------|----------| | Claude Desktop | ⭐⭐⭐⭐⭐ | ⭐⭐ | ⭐⭐⭐⭐⭐ | 💰💰💰 | Required | | Ollama (llama3.2) | ⭐⭐⭐⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐⭐ | Free | Not needed | | Ollama (llama3.1:8b) | ⭐⭐⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐ | Free | Not needed | | LM Studio | ⭐⭐⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐ | Free | Not needed | --- ## 🚀 Quick Start ### Simplest Path (recommended for beginners): 1. **Install Claude Desktop** ```bash # Download from https://claude.ai/download ``` 2. **Configure MCP server** (see README.md) 3. **Start using** - just ask Claude: - "Open Safari" - "What applications are running?" - "Open file ~/Documents/test.txt in TextEdit" ### For Local Models: 1. **Install Ollama** ```bash brew install ollama ollama serve ``` 2. **Load a model** ```bash ollama pull llama3.2 ``` 3. **Configure MCP in Claude Desktop** (see above) 4. **Use local model** through Claude Desktop --- ## 📝 Usage Examples After configuring any of these options, you'll be able to use LLM to manage applications: ``` User: "Open Safari and go to google.com" LLM: [uses open_application and run_applescript] User: "Show me all running applications" LLM: [uses get_running_applications] User: "Create a new document in TextEdit and write 'Hello World'" LLM: [uses open_application and run_applescript] ``` --- ## 🔍 Finding Other MCP Clients If you want to find other options: - [MCP Servers List](https://github.com/modelcontextprotocol/servers) - [MCP Market](https://mcplist.ru) - Official MCP documentation --- ## ⚠️ Important Notes 1. **Compatibility**: Your MCP server works with any MCP-compatible client 2. **Multiple Servers**: You can use multiple MCP servers simultaneously 3. **Privacy**: Local models (Ollama, LM Studio) provide complete privacy 4. **Performance**: Local models require sufficient RAM (16GB+ recommended) --- **Main point**: Choose the client that suits you. The MCP server will work the same with all of them! 🎉