MCP External Memory

What it does

This MCP server provides LLMs with a long-term semantic memory system. It allows the AI to store facts, preferences, and context in a persistent SQLite database and retrieve them later using vector-based similarity searches, overcoming the limitations of short-term context windows.

Tools

• memory_store: Persist text with optional namespaces, tags, and metadata.
• memory_search: Perform semantic search (cosine similarity) over all stored memories.
• memory_get: Retrieve a specific memory by its unique ID.
• memory_delete: Remove a memory from the store by ID.
• memory_list: List memories with optional namespace or tag filtering and pagination.
• memory_update: Modify the content of an existing memory.
• memory_stats: Get overall statistics, including memory count and DB path.

Installation

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "mcp-external-memory": {
      "command": "python3",
      "args": ["-m", "mcp_external_memory"]
    }
  }
}

Supported hosts

• Claude Desktop