MCP Console Automation Server

# Persistent MCP Server Solution ## Problem Solved The MCP server was disconnecting frequently, requiring manual reconnection through the MCP menu. This was disruptive to workflow and prevented continuous automation. ## Solution Implemented Created a persistent MCP server (`persistent-server.ts`) with: ### Key Features 1. **Automatic Reconnection**: Detects disconnections and automatically reconnects with exponential backoff 2. **Keepalive Heartbeat**: Sends periodic pings every 15 seconds to maintain connection 3. **Connection State Management**: Tracks connection state (connected/disconnected/reconnecting) 4. **Graceful Error Recovery**: Handles non-critical errors without crashing 5. **Session Persistence**: Maintains sessions across reconnections 6. **Health Monitoring**: Regular system health checks every 30 seconds ### Technical Details - Extends the base `ConsoleAutomationServer` with persistence features - Overrides process exit handlers to prevent unexpected shutdowns - Implements exponential backoff for reconnection (1s → 2s → 4s → ... max 30s) - Maximum of 10 reconnection attempts before requiring manual intervention - Distinguishes between critical errors (requiring restart) and recoverable errors ## Installation ### Windows (PowerShell) ```powershell .\install-mcp-persistent.ps1 ``` ### Linux/macOS ```bash ./install-mcp-persistent.sh ``` ### Manual Installation Update your Claude configuration file: - Windows: `%APPDATA%\Claude\claude_desktop_config.json` - macOS/Linux: `~/.config/Claude/claude_desktop_config.json` Replace the console-automation server configuration with: ```json "console-automation": { "command": "npx", "args": [ "tsx", "C:\\Users\\yolan\\source\\repos\\mcp-console-automation\\src\\mcp\\persistent-server.ts" ], "env": { "LOG_LEVEL": "info", "NODE_ENV": "production", "NODE_OPTIONS": "--max-old-space-size=4096" } } ``` ## Testing Run the persistent server directly: ```bash npm run start:persistent ``` ## Benefits - **No more manual reconnections**: Server maintains connection automatically - **Improved reliability**: Handles network interruptions gracefully - **Better performance**: Keepalive prevents idle disconnections - **Error resilience**: Recovers from non-critical errors without user intervention - **Session continuity**: Active sessions preserved across reconnections ## Architecture ``` PersistentMCPServer ├── Connection Management │ ├── Keepalive heartbeat (15s interval) │ ├── Connection state tracking │ └── Auto-reconnection with backoff ├── Error Handling │ ├── Critical error detection │ ├── Recovery strategies │ └── Graceful degradation ├── Health Monitoring │ ├── System health checks (30s interval) │ ├── Session health tracking │ └── Performance metrics └── Cleanup & Shutdown ├── Graceful session closure ├── Resource cleanup └── State preservation ``` ## Logs The persistent server provides detailed logging: - Connection state changes - Reconnection attempts - Health check results - Error recovery actions - Active session counts Monitor logs with: ```bash LOG_LEVEL=debug npm run start:persistent ``` ## Troubleshooting ### Server not reconnecting - Check maximum reconnection attempts (default: 10) - Review logs for critical errors - Verify network connectivity ### High memory usage - Adjust `NODE_OPTIONS`: `--max-old-space-size=2048` (lower value) - Check for memory leaks in active sessions ### Connection drops frequently - Reduce keepalive interval (currently 15s) - Check network stability - Review firewall/proxy settings ## Future Improvements - WebSocket support for more stable connections - Connection pooling for multiple Claude instances - Metrics dashboard for monitoring - Automatic session recovery with state preservation - Cloud-based relay for enhanced reliability