Knowledge Graph Memory Server

You have a persistent memory service that uses a local knowledge graph to remember user information across chats. It adds fuzzy searching for quick recall, so you can search for user details and context even when wording isn’t exact.

How to use

Start your memory server and connect it to your MCP client. Your memory server stores entities, relations, and observations in a graph, and you can search for relevant data using fuzzy matching. Begin each interaction by retrieving from memory as needed, then update memory with new facts or changes.

Key actions you can perform through your MCP client include creating entities, adding observations, establishing relationships, updating facts, and querying the graph. Use fuzzy search to discover existing entities before creating duplicates. Retrieve exact nodes when you know their names to inspect their details and relationships. Remove obsolete facts or connections as needed to keep memory accurate.

How to install

Prerequisites you need before starting are: Node.js and npm installed on your system, and optionally Docker if you plan to run a containerized setup.

Install and run the memory MCP server locally by using the following steps.

# Install dependencies for the memory MCP server
npm install

# Start the memory MCP server via the official MCP entrypoint
npx -y github:flrngel/fuzzy-memory-mcp#main

Configuration and environment

You can customize where the memory is stored by setting an environment variable when you run the server. This keeps memory data persistent across restarts.

Example configuration for the client or launcher that points to a custom memory file path.

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": [
        "-y",
        "github:flrngel/fuzzy-memory-mcp#main"
      ],
      "env": {
        "MEMORY_FILE_PATH": "/path/to/custom/memory.json"
      }
    }
  }
}

Notes on usage and behavior

• The memory server uses a local JSON-based storage file by default. You can override this path with MEMORY_FILE_PATH to control where memory is saved.

• Observations should be atomic and stored as distinct facts. You can add, update, or remove observations without affecting unrelated data.

Security and maintenance

Since memory data is stored locally, ensure you apply appropriate access controls to the host machine. Regularly back up the memory file to prevent data loss.

Troubleshooting

If memory data isn’t appearing as expected, verify the MEMORY_FILE_PATH points to a valid location with read and write permissions. Check that your MCP client is connected to the memory server and that the server process is running.

Available tools

search_nodes

Fuzzy semantic search across entity names, types, and observations with a confidence score to help you discover relevant memory data before creating or updating entities.

create_entities

Create new entities with a unique name, type, and initial observations to establish memory of a person, organization, or concept.

add_observations

Add new atomic observations to existing entities to enrich memory without duplicating facts.

create_relations

Link two existing entities with a directed relation in active voice to express how they relate to each other.

open_nodes

Retrieve specific entities by exact name along with their relations to inspect memory connections precisely.

delete_observations

Remove specific observations from an entity to reflect updated reality.

delete_relations

Delete a relation between two entities while preserving the entities themselves.

delete_entities

Permanently remove an entity and all its connections from memory.

read_graph

Retrieve the entire memory graph for offline analysis or debugging.

read_graph

Retrieve the entire memory graph for offline analysis or debugging.