MemoryGraph is a graph-based Model Context Protocol (MCP) server that gives AI coding agents persistent memory. Store patterns, track relationships, retrieve knowledge across sessions.
MemoryGraph is a graph-based Model Context Protocol (MCP) server that provides persistent memory functionality for AI coding agents. By storing knowledge patterns, tracking complex relationships, and enabling intelligent cross-session retrieval, it effectively overcomes the limitations of traditional vector search methods in handling long-term and temporally dependent tasks.
The core strength of MemoryGraph lies in its use of a graph structure to capture diverse relationships—such as causal, solution-based, and contextual connections—between entities, moving beyond simple flattened storage. This enables deep associative queries into past decisions and learned experiences, along with enhanced contextual understanding.
MemoryGraph is compatible with various MCP clients, such as Claude Code, and offers two operational modes: "Core" and "Extended." The Core mode focuses on daily memory storage and fuzzy recall, while the Extended mode adds capabilities like database statistics and complex relationship querying.
Users can leverage its memory tools by configuring AI agents (e.g., via CLAUDE.md) and providing explicit prompts. It supports multiple memory types (e.g., solutions, problems, code patterns) and relationship creation, and offers a range of backend options—including SQLite, FalkorDBLite, FalkorDB, Neo4j, and Memgraph—to accommodate different deployment and performance requirements.