Memory

This Memory MCP Server provides a persistent knowledge graph that stores entities, relations, observations, and lessons. It helps you remember user details across interactions, learn from past errors via a lesson system, and offer contextual recommendations to improve responses over time.

How to use

You interact with the Memory MCP Server through your MCP client by starting a local server process or running a container, then connecting your client to it. The server stores information as entities, their relationships, and observations. It can also manage lessons that capture common errors and their validated solutions. Use it to remember user preferences, topics of interest, and recurring problems to improve future conversations.

How to install

Prerequisites: you need Node.js and npm or a container runtime like Docker installed on your machine.

# Option A: Run Memory MCP Server via Docker (recommended for quick setup)
docker run -i -v claude-memory:/app/dist --rm mcp/memory

Additional configuration and usage notes

You can run multiple MCP server configurations to connect through different methods. The following options show how to start the Memory MCP Server locally and via a container, including an example environment variable to customize storage paths.

Server connection configurations

The Memory MCP Server can be started in stdio mode (local process) using either Docker or NPX. The configurations shown reflect explicit command invocations you can use to start the server and then connect your Cursor/Claude client.

Security and maintenance notes

Keep memory data secure by restricting access to the storage volume or memory file path. Regularly back up memory.json and lesson.json, and rotate access credentials if you expose the memory server to a shared environment.

Troubleshooting tips

If you cannot connect from the client, verify the server process is running and listening on the expected stdio interface. Check for permission issues on the memory storage path and ensure that environment variables used to customize paths are set in your startup command.

Example use cases

• Remember user preferences and recurring topics to tailor responses. - Track and apply lessons from past errors to improve future outcomes. - Retrieve targeted observations and relationships to provide contextual memory during chats.

Available tools

create_entities

Create multiple new entities in the knowledge graph. Each entity has a name, type, and a list of observations.

create_relations

Create multiple relations between entities. Each relation connects a source to a target with a specific relation type.

add_observations

Add new observations to existing entities. Returns added observations per entity and fails if the entity does not exist.

delete_entities

Remove entities and their relations. Performs cascading deletion of related edges.

delete_observations

Remove specific observations from entities. Silent if the observation does not exist.

delete_relations

Remove specific relations from the graph. Silent if the relation does not exist.

read_graph

Read and return the entire knowledge graph, including all entities and relations.

search_nodes

Search for nodes by a query across names, types, and observations and return matching entities with their relations.

open_nodes

Retrieve specific nodes by name and return the requested entities along with the relations between them.

create_lesson

Create a new lesson from an error and its solution, with validation and metadata tracking.

find_similar_errors

Find similar errors and their solutions using fuzzy matching and return matching lessons sorted by success rate.

update_lesson_success

Update success tracking for a lesson, adjusting success rate and frequency, and recording last update time.

get_lesson_recommendations

Get relevant lessons for a given context, sorted by relevance and success rate, including full solution details.