# MCP Server Configuration

## Running MCP Servers

### Option 1: Directly with Python
```bash
cd mcp
pip install -r requirements.txt
python homelab.py
```

### Option 2: Via Claude Desktop Config

Add to your Claude Desktop config (`~/Library/Application Support/Claude/claude_desktop_config.json` on macOS or `%APPDATA%\Claude\claude_desktop_config.json` on Windows):

```json
{
  "mcpServers": {
    "homelab": {
      "command": "python",
      "args": ["/path/to/ai-skills-api/mcp/homelab.py"],
      "env": {
        "DOCKER_HOST": "unix:///var/run/docker.sock"
      }
    },
    "gameservers": {
      "command": "python",
      "args": ["/path/to/ai-skills-api/mcp/gameservers.py"],
      "env": {
        "GAME_SERVERS_DIR": "/opt/game-servers"
      }
    },
    "skills": {
      "command": "python",
      "args": ["/path/to/ai-skills-api/mcp/skills.py"],
      "env": {
        "SKILLS_API_URL": "http://localhost:8080"
      }
    }
  }
}
```

## Available Tools

### homelab
- `container_status` - Get Docker container status
- `list_containers` - List all containers
- `start_container` - Start a container
- `stop_container` - Stop a container
- `restart_container` - Restart a container
- `container_logs` - Get container logs
- `system_resources` - Get CPU/memory/disk usage
- `run_command` - Run shell command (use carefully)
- `docker_compose_action` - Run docker-compose actions

### gameservers
- `list_servers` - List all game servers
- `get_server_config` - Get server config
- `update_server_config` - Update server config
- `server_status` - Get server status
- `start_server` - Start a game server
- `stop_server` - Stop a game server
- `get_server_logs` - Get server logs
- `create_server` - Create new game server
- `delete_server` - Delete a game server
- `get_templates` - Get available templates

### skills
- `get_skill` - Get skill by ID
- `search_skills` - Search skills
- `list_skills` - List skills
- `get_context` - Get context bundle
- `get_conventions` - Get conventions
- `get_snippets` - Get snippets
- `check_cache` - Check response cache
- `get_memory` - Get project memory
- `add_memory` - Add project memory
- `create_skill` - Create new skill

## Token Savings Pattern

When using agents:

1. **Before asking**: Call `skills/check_cache` with your prompt
2. **If cached**: Use the cached response directly
3. **If not cached**: Call `skills/get_context` to inject relevant skills/conventions
4. **After response**: Optionally call `skills/add_memory` to save important decisions

This pattern avoids:
- Re-sending your coding standards every request
- Re-explaining project architecture
- Re-asking questions you've asked before