Memory-Graph – Knowledge Graph Memory for Claude Code with SQLite/Neo4j/Memgraph (github.com)

MemoryGraph is a new Model Context Protocol (MCP) server that gives AI coding agents persistent, graph-structured memory—originally optimized for Claude Code but compatible with any MCP-enabled client (Cursor, Continue, etc.). Installation is simple (pip install memorygraphMCP) and you enable it by pointing your MCP client at the memorygraph command; by default data is stored at ~/.memorygraph/memory.db. It supports multiple backends (SQLite, Neo4j, Memgraph), runs in lite → standard → full profiles, and exposes tools like store_memory, search_memories and create_relationship to let agents persist and query development patterns across sessions and projects. Significance: unlike flat CLAUDE.md files or plain vector stores, MemoryGraph models explicit relationships (seven categories such as CAUSES, SOLVES, APPLIES_TO, SIMILAR_TO) so agents can retrieve causal chains, pattern variants, and workflow dependencies—e.g., a query about “retry logic” can return the full causal chain from bug → cause → fix. Technical features include semantic/graph search, pattern recognition, session briefings, graph analytics, token-limited context retrieval, and 44 specialized tools in full mode. Building on the open MCP spec avoids vendor lock-in: you can migrate memory between MCP servers or run hybrid setups (local SQLite for quick use, Neo4j/Memgraph for advanced analytics).