Best MCP servers for knowledge graphs
Some questions are about relationships, not records: who is connected to whom, what depends on what, how does this entity reach that one through a chain of links. Knowledge graphs model the world as nodes and edges, and an agent that can query a graph can answer multi-hop questions, traverse dependencies, and reason over structure that flat tables flatten away. The right server depends on your substrate, a dedicated graph database, a multi-model store that does graphs alongside other shapes, or a lightweight entity-relation memory, but the recurring need is the same: let the agent run graph queries and walk relationships directly. The servers below cover the common shapes, each a real MCP server with a verified, current install config.
Neo4j
Neo4j
Neo4j's official MCP server lets agents introspect a graph schema and run read or write Cypher against any Neo4j deployment.
Neo4j's server lets an agent run Cypher queries against a native property graph, the most direct option when your knowledge lives in a purpose-built graph database and you need real traversal and pattern matching.
Neo4j for knowledge graphs →Memory
Anthropic (reference)
Reference MCP server giving an agent a persistent local knowledge graph of entities, relations, and observations.
The reference Memory server maintains a local knowledge graph of entities, relations, and observations, a lightweight way to give an agent persistent graph-shaped memory without standing up a database.
Memory for knowledge graphs →SurrealDB
SurrealDB
SurrealDB's official MCP server lets agents query and mutate a SurrealDB instance with SurrealQL and manage SurrealDB Cloud.
SurrealDB's server exposes a multi-model database that handles graph relations alongside documents and tables, ideal when your data is part-graph, part-relational and you want one store for both.
SurrealDB for knowledge graphs →