Enhancement: DXT (Desktop Extensions) packaging support

[bm-claudeai]

Summary

Investigate implementing DXT (Desktop Extensions) packaging for Basic Memory to enable one-click installation in Claude Desktop and other MCP-compatible applications.

Background

Anthropic released the DXT specification on June 26, 2024 to enable packaging MCP servers as single-click installable extensions (.dxt files). This would significantly improve Basic Memory's user experience by eliminating manual installation and configuration steps.

Current Status: NOT READY FOR IMPLEMENTATION

After investigation and implementation attempts, DXT is too immature for production use with Basic Memory:

Major Blocking Issues

1. Binary Permissions Problem (anthropics/dxt#12)

   • Binary files lose executable permissions when packaged
   • Causes EACCES errors during execution
   • No current solution beyond manual chmod after installation

2. Platform-Specific Dependencies (anthropics/dxt#17)

   • Basic Memory has native binary dependencies (pydantic_core, SQLite, Pillow, etc.)
   • No guidance for cross-platform Python packages with native extensions
   • Community consensus: need separate packages for each platform/architecture

3. Python Path Issues

   • Claude Desktop can't find python command due to PATH limitations
   • python3 command has better compatibility but still platform-dependent

4. DXT Immaturity

   • Version 0.1.0 explicitly labeled as preview/work in progress
   • Anthropic states they're "looking forward to working with the greater community on evolving and changing the format"
   • Multiple critical issues unresolved

Implementation Attempt Details

We successfully created a working DXT package structure:

• Created comprehensive manifest.json with all 19 Basic Memory MCP tools
• Implemented dependency bundling following Anthropic's patterns
• Added automated build process via justfile (just dxt)
• Package size: ~30MB with bundled dependencies

However, encountered these runtime issues:

ModuleNotFoundError: No module named 'pydantic_core._pydantic_core'

The native binary modules are present but not loading correctly in Claude Desktop's execution environment.

Basic Memory Dependencies Requiring Native Binaries

• pydantic → pydantic_core (Rust extensions)
• sqlalchemy → SQLite C extensions
• pillow → Image processing libraries
• aiosqlite → SQLite async bindings
• watchfiles → File system monitoring (Rust)

Recommendation

Wait for DXT v0.2+ before implementing. Focus on current distribution channels:

• PyPI (uv tool install basic-memory)
• Homebrew (brew install basicmachines-co/basic-memory/basic-memory)
• Docker containers

Future Implementation Plan

When DXT matures (estimated 3-6 months):

1. Multi-Platform Strategy: Build separate DXT packages for major platforms:

   • basic-memory-macos-arm64.dxt
   • basic-memory-macos-x86_64.dxt
   • basic-memory-windows-x86_64.dxt
   • basic-memory-linux-x86_64.dxt

2. Distribution Infrastructure:

   • Automated GitHub Actions builds for each platform
   • Download page with platform detection
   • Clear installation instructions

3. Testing Protocol:

   • Verify all 19 MCP tools work correctly
   • Cross-platform compatibility testing
   • Performance benchmarking vs. native installation

References

• DXT Specification
• Anthropic's DXT Announcement
• DXT GitHub Issues

Branch

Implementation work is available on the dxt-packaging branch for future reference.