RMANOV/sqlite-memory-mcp
SQLite-backed MCP Memory Server with WAL concurrent safety, FTS5 search, session tracking, task management, and cross-machine bridge sync
What's novel
SQLite-backed MCP Memory Server with WAL concurrent safety, FTS5 search, session tracking, task management, and cross-machine bridge sync
Code Analysis
8 files read · 3 roundsA production-grade SQLite-backed MCP memory server that provides a provenance-rich knowledge graph with hybrid BM25+semantic search, cross-machine git-based sync, task management, and multi-agent debate protocols for coding agents like Claude and Codex.
Strengths
Exceptionally deep implementation with measured design decisions (recall budget uses a live FTS5 probe table for tokenizer parity, HLC clocks for cross-machine sync, content shrink guards, enrichment invariants) and a genuinely sophisticated 6-signal multiplicative re-ranking pipeline fused with RRF vector search. The codebase demonstrates real engineering rigor — every non-obvious choice is documented with the measurement that justified it.
Weaknesses
The flat module layout with 60+ top-level files and several 100KB+ modules (db_utils at 345KB, debate at 174KB, task_tray at 141KB) makes navigation and maintenance difficult; the lack of a proper package structure and the inability to verify test quality (138 test files exist but none were readable) are notable gaps. The project's breadth (7 micro-servers, PyQt6 GUI, n8n integration, Obsidian publishing, Notion import) risks feature sprawl beyond the core memory use case.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
138
Contributors
1
Files
37
Active weeks
3
Repository
Language
Python
Stars
2
Forks
0
License
MIT