Memory

Memory is an MCP server that provides a persistent, knowledge graph-based storage layer for AI assistants like Claude across separate sessions. It connects LLM clients directly to a local graph database stored on the filesystem as a JSONL file, bypassing stateless chat limitations. Software developers, researchers, and personal AI users use this server to build continuous contextual awareness for their autonomous agents and chat sessions. The server represents data through three fundamental constructs: entities that act as primary nodes, active-voice relations that define directed connections between nodes, and atomic observations that store specific facts about individual entities. By exposing dedicated tools for creating, updating, searching, and deleting nodes and edges, along with an MCP resource for live graph synchronization, the server enables language models to retain user preferences, organizational hierarchies, project details, and interaction histories indefinitely without requiring external cloud databases.

Category: AI Memory & Context

Tags: context, knowledge-graph, persistent-memory, rag

Visit Memory

How to install and configure Memory

Follow these steps to configure the Memory MCP server for Claude Desktop: 1. Open your claude_desktop_config.json configuration file. 2. Add the Memory server definition under the mcpServers object using npx: json { "mcpServers": { "memory": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-memory"] } } } 3. On Windows platforms, invoke npx via the command shell by setting "command": "cmd" and "args": ["/c", "npx", "-y", "@modelcontextprotocol/server-memory"]. 4. Optionally, configure a custom storage path by adding "env": { "MEMORY_FILE_PATH": "/path/to/custom/memory.jsonl" }. 5. Save the configuration and restart Claude Desktop.

What you can do with Memory

• Storing user identity, contact details, communication styles, and workplace relationships persistently across multiple independent conversations with Claude. * Managing organizational knowledge by linking projects, team members, and internal systems through directed graph relationships and atomic observations. * Querying specific knowledge graph nodes and associated connections using text search across names, types, and observation contents. * Tracking meeting notes, personal preferences, and career goals over time by incrementally adding or removing discrete observation facts. * Exposing graph mutations dynamically to subscribed MCP clients using the live-updating memory://knowledge-graph resource URI.

Key facts

• https://github.com/modelcontextprotocol/servers
• AI Memory & Context, Notes & Knowledge Management
• context, knowledge-graph, persistent-memory, rag

Part of MCP Servers

How do I install Memory?

You can install and run the Memory server by adding @modelcontextprotocol/server-memory to your MCP client configuration using npx, or by running the official Docker container mcp/memory. On Windows systems, configure the execution command through cmd /c.

What can Memory do?

Memory provides tools to create, read, update, and search a persistent local knowledge graph. It stores data as entities, active-voice relationships, and discrete text observations, letting AI clients recall personal information, preferences, and complex domain relationships across chats.

Which MCP clients work with Memory?

Memory works with any client supporting the Model Context Protocol, including Claude Desktop, VS Code, and VS Code Insiders. It connects via standard IO using npx or Docker container execution.

Where is the knowledge graph data stored?

Data is stored locally in a JSONL file. By default, it saves to memory.jsonl in the server directory, but you can specify a custom storage location using the MEMORY_FILE_PATH environment variable.

Is Memory open source?

Yes, the Memory MCP server is open source and distributed under the terms of the MIT License. You are free to use, modify, and distribute the software accordingly.