I built this because Claude Code keeps forgetting what it learned.
Every session I'd re-explain the architecture, re-discover the same bugs, re-learn the same solutions. CLAUDE.md helps but it's manual and doesn't capture relationships between things.
memory-graph is an MCP server that stores entities, observations, and relationships in a knowledge graph. The key insight: coding agents need to remember what solved problems, not just find similar content.
Three backends:
- SQLite (default) - single file, zero config, ~50MB with embeddings
- Neo4j - when you need scale
- Memgraph - when you need speed
The graph relationships (SOLVES, CAUSES, RELATES_TO) let you query "what fixed the JWT bug last time?" instead of just "find auth-related code."
pip install memorygraphMCP
Then add to claude_desktop_config.json:
{"mcpServers": {"memory-graph": {"command": "memory-graph"}}}
Curious what others are doing for persistent memory in coding agents. The official @modelcontextprotocol/server-memory is great but uses JSONL - I wanted queryable relationships.
I built this because Claude Code keeps forgetting what it learned.
Every session I'd re-explain the architecture, re-discover the same bugs, re-learn the same solutions. CLAUDE.md helps but it's manual and doesn't capture relationships between things.
memory-graph is an MCP server that stores entities, observations, and relationships in a knowledge graph. The key insight: coding agents need to remember what solved problems, not just find similar content.
Three backends: - SQLite (default) - single file, zero config, ~50MB with embeddings - Neo4j - when you need scale - Memgraph - when you need speed
30 MCP tools, 409 tests passing (90%+ coverage), 8 ADRs documenting decisions.
The graph relationships (SOLVES, CAUSES, RELATES_TO) let you query "what fixed the JWT bug last time?" instead of just "find auth-related code."
pip install memorygraphMCP
Then add to claude_desktop_config.json: {"mcpServers": {"memory-graph": {"command": "memory-graph"}}}
Curious what others are doing for persistent memory in coding agents. The official @modelcontextprotocol/server-memory is great but uses JSONL - I wanted queryable relationships.
https://github.com/gregorydickson/memory-graph