Memory MCP Server
Persistent knowledge graph memory allowing AI to retain user information across sessions.
Data last scanned yesterday · Reviewed 2 months ago
Overview
The Memory MCP Server functions as a local persistent storage system, enabling AI models to maintain a long-term knowledge graph. It organizes data into entities, observations, and directed relations, providing a structured way to track complex details about people, organizations, or events over time. Developers utilize this server to create, update, or remove specific facts and connections within the graph. Because it supports granular operations like adding individual observations to existing entities or executing cascading deletions, it allows for fine-tuned management of the information stored by your AI assistant.
Our verdict
This server receives a security score of 76/100, placing it at a medium risk level. The most significant concern is the absence of a read-only mode, as every installation grants full write access to the underlying knowledge graph. While it benefits from being actively maintained by the official Model Context Protocol organization and has a minimal footprint with only one direct dependency, users should exercise caution given the potential for broad write operations.
- Maintained by the official Model Context Protocol organization
- Active development with updates occurring daily
- Extremely minimal dependency footprint with only one direct library
- Large community footprint with 89,054 GitHub stars
- No read-only mode restricts permission granularity
- Authentication methods currently lack formal security review
- License status is marked as NOASSERTION
Setup
{
"mcpServers": {
"memory": {
"args": [
"-y",
"@modelcontextprotocol/server-memory"
],
"command": "npx"
}
}
}Paste into ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows), then restart Claude Desktop. Some servers require API keys or tokens as environment variables — check the project's README and add an "env" object if needed.
- No read-only mode — running this grants full read/write access.
- No built-in authentication — anyone with access to this config can invoke every tool it exposes.
Tools
| Tool | Description | Risk |
|---|---|---|
| create_entities | Create multiple new entities in the knowledge graph | medium |
| create_relations | Create multiple new relations between entities | medium |
| add_observations | Add new observations to existing entities | medium |
| delete_entities | Remove entities and their relations | high |
| delete_observations | Remove specific observations from entities | high |
| delete_relations | Remove specific relations from the graph | high |
| read_graph | Read the entire knowledge graph | low |
| search_nodes | Search for nodes based on query | low |
| open_nodes | Retrieve specific nodes by name | low |
Compatibility
| Client | Local | Docker | Remote | Read-only |
|---|---|---|---|---|
| Claude Desktop | ||||
| Cursor | ||||
| VS Code | ||||
| Windsurf | ||||
| ChatGPT |
Frequently asked questions
›How can I customize where my memory data is saved?
You can change the storage location by setting the MEMORY_FILE_PATH environment variable to your desired path in your configuration file.
›Does deleting an entity remove its associated relations?
Yes, deleting an entity triggers a cascading deletion of all relations associated with that entity.
›How are new memory changes communicated to the client?
The server automatically emits notifications/resources/updated events whenever mutation tools modify the knowledge graph.
›Can I use this server on Windows?
Yes, you should use cmd /c to launch the npx command within your Claude desktop configuration file.
›What happens if I try to create an entity that already exists?
The create_entities tool will ignore any entities that have already been created with an existing identifier.
Alternatives
Badge
Maintain this server? Add the live badge to your README.