memorygraph

Entity-anchored, vector-searchable, salience-weighted graph memory for Claude Code.

MemoryGraph is an MCP (Model Context Protocol) server that gives Claude Code persistent, intelligent memory. It stores information anchored to named entities, finds relevant memories through semantic search, and automatically discovers clusters of related entities using Louvain community detection.

Everything runs locally. No cloud. No API keys. Your data stays on your machine.

claude mcp add --scope user memory-graph -- npx @memorygraph/server

---

Claude Code has no long-term memory between conversations. Each session starts from zero. MemoryGraph fixes this by providing a persistent memory layer that:

• Remembers across sessions — Store facts, decisions, preferences, and context that persist between conversations
• Finds what matters — Semantic vector search finds relevant memories by meaning, not just keywords
• Prioritises important memories — A salience system ensures critical information surfaces first while stale information fades naturally
• Discovers connections — Louvain community detection reveals hidden relationships between entities
• Grows smarter over time — Memories that get accessed more often become stronger; unused ones decay gracefully

| Benefit | How | |---------|-----| | Context continuity | Memories persist across Claude Code sessions — no repeated explanations | | Semantic recall | Vector search understands meaning: "deployment issues" finds memories about "CI/CD failures" | | Automatic prioritisation | Salience scoring surfaces important memories first, decays irrelevant ones | | Entity graph | Named entities (people, projects, tools) form a knowledge graph with typed relations | | Associative recall | Cluster expansion finds related memories through entity communities | | Deduplication | Content hashing prevents duplicate memories; entity