doobidoo/mcp-memory-service v4.3.2

v4.3.2 - Repository Organization & PyTorch Download Fix

on GitHub

🎯 Repository Organization & PyTorch Download Fix

🔥 Major Achievements

• ✅ Fixed PyTorch Download Issue: Completely eliminated the 230MB+ PyTorch downloads that plagued Claude Desktop on every startup
• ✅ Repository Cleanup: Professional organization with clean root directory and logical documentation hierarchy
• ✅ Performance Boost: Reduced Claude Desktop startup time from ~60 seconds to ~3 seconds for Windows users

🛠️ Technical Fixes

PyTorch Download Resolution:

• Root Cause: UV package manager isolation prevented offline environment variables from taking effect
• Solution: Created scripts/memory_offline.py launcher that sets offline mode BEFORE any ML library imports
• Implementation: Updated Claude Desktop config to use direct Python execution instead of UV
• Result: Complete elimination of repeated PyTorch downloads using cached models

Environment & Configuration:

• Fixed UV environment variable inheritance issues
• Added comprehensive HuggingFace offline configuration (HF_HUB_OFFLINE, TRANSFORMERS_OFFLINE)
• Implemented fallback offline setup in core module initialization
• Optimized Windows cache path management for sentence-transformers

📁 Repository Structure Overhaul

Documentation Organization:

• docs/guides/ - User guides and setup instructions
• docs/technical/ - Technical documentation and implementation details
• docs/deployment/ - Production deployment guides
• docs/installation/ - Installation and setup guides
• docs/integrations/ - Third-party integrations (Gemini, etc.)

Cleanup & Consolidation:

• Moved service management scripts to /scripts directory
• Removed obsolete debug scripts and development notes
• Consolidated duplicate setup guides
• Eliminated cluttered root directory files

📊 Impact & Benefits

• ⚡ Startup Performance: 95% faster Claude Desktop initialization
• 💾 Bandwidth Savings: No more 230MB downloads on every startup
• 🏢 Enterprise Ready: Professional repository structure
• 🛡️ Reliability: Stable offline operation with cached models
• 📚 Maintainability: Organized documentation and consolidated scripts

🔄 Migration Guide

For Existing Users:

1. Update your Claude Desktop config to use the new scripts/memory_offline.py launcher
2. Use the updated examples/claude_desktop_config_windows.json template
3. Documentation moved - check the new /docs structure for guides

New Windows Template:

{
  "mcpServers": {
    "memory": {
      "command": "python",
      "args": ["path/to/mcp-memory-service/scripts/memory_offline.py"],
      "env": {
        "MCP_MEMORY_STORAGE_BACKEND": "sqlite_vec",
        "HF_HUB_OFFLINE": "1",
        "TRANSFORMERS_OFFLINE": "1"
      }
    }
  }
}

🏗️ Files Changed

• New: scripts/memory_offline.py - Offline mode launcher
• New: PYTORCH_DOWNLOAD_FIX.md - Detailed fix documentation
• New: examples/claude_desktop_config_windows.json - Windows template
• Moved: Documentation files reorganized to /docs subdirectories
• Updated: pyproject.toml version 4.3.0 → 4.3.2
• Enhanced: CHANGELOG with comprehensive release notes

This release solves the persistent PyTorch download issue that has been frustrating Windows users and establishes a professional, enterprise-ready repository structure.

The stubborn PyTorch download problem is now completely resolved! 🎉