FCCS MCP Agentic Server

# Feedback Service Fix ## Issue The feedback service was throwing an error: `unsupported operand type(s) for +=: 'NoneType' and 'int'` when trying to update metrics. ## Root Cause The error occurred when: 1. Database records had NULL values in integer columns (total_calls, success_count, etc.) 2. SQLAlchemy tried to perform `+=` operations on these NULL values 3. Using the same session for both execution logging and metrics updates caused conflicts ## Solution The fix includes: 1. **Separate Sessions**: Metrics updates now use a separate database session to avoid conflicts 2. **Explicit None Handling**: All database values are checked for None and converted to defaults before use 3. **Assignment Instead of +=**: Changed from `metrics.total_calls += 1` to `metrics.total_calls = new_total` to avoid None issues 4. **Better Error Handling**: Added try-except blocks to prevent feedback service errors from breaking tool execution ## Changes Made ### `fccs_agent/services/feedback_service.py` - Renamed `_update_metrics()` to `_update_metrics_separate_session()` - Uses a separate session for metrics updates - Explicitly handles None values before calculations - Uses assignment operations instead of += ### `fccs_agent/agent.py` - Added error handling to prevent feedback service errors from breaking tool execution - Feedback service initialization failures are now non-fatal ## Testing Run the test script to verify the fix: ```bash python scripts/test_feedback_service.py ``` ## Applying the Fix **Important**: If you're using the MCP server, you need to restart it to pick up the changes: 1. Stop the MCP server (if running) 2. Clear Python cache (optional but recommended): ```bash find . -type d -name __pycache__ -exec rm -r {} + find . -type f -name "*.pyc" -delete ``` 3. Restart the MCP server: ```bash python -m cli.mcp_server ``` Or if using Claude Desktop, restart Claude Desktop to reload the MCP server. ## Verification After restarting, test with: ```bash python scripts/test_feedback_service.py ``` The test should show: - [OK] Feedback service initialized - [OK] Execution logged - [OK] Metrics retrieved - [OK] Multiple executions logged - [OK] Updated metrics