Knowledge Graph Memory Server

What it does

The memory server stores facts in a local knowledge graph so a model can retain information across chats. Entities have a unique name, a type, and a list of observations. Relations are directed connections written in active voice, and observations are single facts attached to an entity.

Data persists in a JSONL file. The default location is memory.jsonl in the server directory, and MEMORY_FILE_PATH can point the server at a custom file. A knowledge-graph resource exposes the full graph as JSON and sends update notifications when mutation tools change it.

The server runs with npx or Docker. The Docker example mounts a named volume so the memory file survives container replacement.

Configuration

Run with npx -y @modelcontextprotocol/server-memory, optionally setting MEMORY_FILE_PATH to choose the storage file.

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"],
      "env": {"MEMORY_FILE_PATH": "/path/to/custom/memory.jsonl"}
    }
  }
}

Source: https://github.com/modelcontextprotocol/servers/tree/main/src/memory. Placeholders only. Replace them in your own environment, and never commit real keys to a repository.

Verified tools and features

• create_entities and create_relations Config checked

  Creates entities with types and observations, and directed relations between existing entities.

• add_observations Config checked

  Adds atomic fact strings to an existing entity and returns what was added.

• search_nodes and open_nodes Config checked

  Searches names, types and observation text, or retrieves named nodes with relations between them.

• knowledge-graph resource Config checked

  Exposes the full graph at memory://knowledge-graph as application/json, with update notifications on mutations.

Security notes