MCP Memory Server

What is MCP Memory Server?

MCP Memory Server is a Model Context Protocol (MCP) server written in Go that provides persistent memory for Large Language Models. It stores concepts, code snippets, notes, and any information you want your LLM to remember across conversations, keeping all data locally on your machine.

How to use MCP Memory Server?

Clone the repository, build the Go binary, then configure Claude Desktop by adding the server’s absolute binary path and the MCP_DATA_DIR environment variable to the MCP configuration file. Restart Claude Desktop to use commands such as “Remember this: …”, search queries, category listings, and statistics requests. The server exposes five tools: remember, recall, forget, list_memories, and memory_stats.

Key features of MCP Memory Server

• Persistent memory across LLM sessions
• Smart organization with categories and tags
• Natural language search for retrieval
• Usage analytics (most‑accessed memories)
• Local file‑based storage with no external dependencies
• Fast in‑memory indexing for quick access
• Environment variable based configuration
• Thread‑safe concurrent access with proper locking

Use cases of MCP Memory Server

• Retain project context (architecture, tools, decisions) across LLM conversations
• Store and quickly retrieve code snippets or documentation
• Organise research notes by category and tags for later recall
• Track frequently accessed information with usage statistics
• Keep personal reference data accessible without online services

FAQ from MCP Memory Server

How are memories stored?

Each memory is saved as a JSON file in the configured MCP_DATA_DIR directory (default ~/.mcp-memory/memories/). Every entry includes a unique content‑based ID (SHA256 hash), content, optional summary, categories, tags, timestamps, access statistics, and custom metadata.

What tools does it provide to Claude?

Five tools: remember (store new information), recall (search stored memories), forget (delete by ID), list_memories (list with filtering), and memory_stats (usage statistics). Each tool has specific required and optional parameters.

What are the server’s dependencies and runtime requirements?

Go 1.19 or later and the Claude Desktop application. No internet connection is required for basic operation; all data stays local.

Can I configure search limits and log settings?

Yes. Environment variables control log level (MCP_LOG_LEVEL), log format (MCP_LOG_FORMAT), maximum result count (MCP_MAX_RESULTS, default 20), and maximum memory file size (MCP_MAX_FILE_SIZE, default 100 MB).

Does the server support semantic search?

Semantic search via vector embeddings is planned but not yet implemented. The MCP_ENABLE_EMBEDDINGS and MCP_EMBEDDING_MODEL environment variables exist as future placeholders.