mcp-memory-service is a production-ready MCP server that gives AI assistants persistent, searchable memory. It uses a hybrid backend combining fast local reads (around 5ms) via SQLite-vec with ONNX embeddings and background synchronization to Cloudflare, so memory stays available offline while staying in sync across devices and clients.

Use it when you want an AI assistant to retain context across sessions - automatically injecting relevant past memories, surfacing the top-3 most relevant memories at session start with relevance scores, and supporting concurrent access from multiple AI applications without database locking. It supports 13+ AI applications including Claude Desktop, VS Code, Cursor, and Continue, and offers team collaboration through OAuth 2.1. Do not use it if you need a fully offline-only setup with no cloud component at all - the hybrid backend's sync layer relies on Cloudflare, though a pure sqlite backend option exists if cloud sync isn't wanted.

• Zero database locking, allowing concurrent access from multiple clients at once.
• Hybrid storage backend: fast local reads plus background synchronization through Cloudflare.
• Natural Memory Triggers with 85%+ accuracy for automatic, context-aware memory retrieval.
• Intelligent memory search with automatic context injection into conversations.
• Team collaboration via OAuth 2.1.
• Memory consolidation system with decay scoring, inspired by how memory consolidation works during sleep.
• Multi-client support across Claude Desktop, VS Code, Cursor, Continue, and other MCP-compatible applications.
• Offline-capable via SQLite-vec with local ONNX embeddings.
• Git-aware context integration.

pip install mcp-memory-service
# or
uv pip install mcp-memory-service
# or from source
git clone https://github.com/doobidoo/mcp-memory-service.git
cd mcp-memory-service && python install.py
# or via Docker
docker-compose up -d
# or via Smithery for Claude
npx -y @smithery/cli install @doobidoo/mcp-memory-service --client claude

Configure Claude Desktop with:

{
  "mcpServers": {
    "memory": {
      "command": "memory",
      "args": ["server"],
      "env": { "MCP_MEMORY_STORAGE_BACKEND": "hybrid" }
    }
  }
}

MCP_MEMORY_STORAGE_BACKEND can be set to hybrid, cloudflare, or sqlite. Requires Python 3.12+ (check sqlite-vec compatibility on 3.13); macOS users may need Homebrew Python for SQLite extension support. Initial setup downloads a small (~25MB) embedding model automatically.

AI application builders and power users who want persistent, cross-session memory shared across many MCP clients, with fast local access and optional cloud sync for team or multi-device use.