doobidoo/mcp-memory-service v8.71.0

v8.71.0 - Memory Management APIs and Graceful Shutdown

on GitHub

Memory Management APIs and Graceful Shutdown

This release addresses GitHub Discussion #331 where users reported orphaned MCP sessions consuming excessive memory.

New Features

Cache Cleanup Functions

• clear_all_caches() - Clear storage and service caches
• get_memory_usage() - Get process memory statistics (RSS, VMS, available memory)
• get_cache_stats() - Get cache hit/miss statistics
• clear_model_caches() - Clear embedding model caches
• get_model_cache_stats() - Get model cache statistics

Graceful Shutdown

• _cleanup_on_shutdown() - Cleanup function for signal handlers
• Updated shutdown() method with proper cache and connection cleanup
• Added atexit handler for normal process exit
• SIGTERM and SIGINT handlers now call cleanup before exit

Memory Management API Endpoints

• GET /api/memory-stats - Get detailed memory usage (process memory + cache stats + model cache stats)
• POST /api/clear-caches - Clear all caches manually (returns memory freed)

Documentation

• New file: docs/troubleshooting/memory-management.md - Comprehensive guide for monitoring and managing memory

Benefits

• Prevents memory leaks from orphaned sessions
• Enables proactive memory monitoring and cleanup
• Graceful resource release on server shutdown
• Production-ready memory management for long-running deployments

Installation

pip install --upgrade mcp-memory-service

Or with uv:

uv pip install --upgrade mcp-memory-service

Related

• Fixes: GitHub Discussion #331
• PR: #331

Full Changelog: v8.70.0...v8.71.0