Super-Memory

A semantic memory storage and retrieval MCP (Model Context Protocol) server using LanceDB and sentence transformers.

What is Super-Memory?

Super-Memory gives your AI agents long-term memory across sessions. It stores and retrieves information using semantic embeddings, so agents can recall relevant context from previous conversations, files, and web pages.

Features

• Semantic search - Query memories by meaning, not just keywords
• File memory - Read and store local file contents
• Web memory - Fetch and store web page contents
• Boomerang context - Special support for Boomerang Protocol session state
• Local storage - All data stays on your machine in ./memory_data

Tools

Installation

Using uv (recommended)

uv tool install super-memory-mcp

Using pip

pip install super-memory-mcp

Manual / Development

git clone https://github.com/Veedubin/Super-Memory.git
cd Super-Memory
uv sync
uv run super-memory-mcp

OpenCode Configuration

Add to your .opencode/opencode.json:

{
  "mcp": {
    "super-memory-mcp": {
      "type": "local",
      "command": ["uv", "run", "super-memory-mcp"],
      "enabled": true
    }
  }
}

Or if installed with uv tool:

{
  "mcp": {
    "super-memory-mcp": {
      "type": "local",
      "command": ["super-memory-mcp"],
      "enabled": true
    }
  }
}

Requirements

• Python >= 3.13
• CUDA (optional but recommended) - falls back to CPU automatically
• ~500MB disk space for the embedding model (downloaded on first run)

First Run

On first startup, Super-Memory will download the sentence-transformers/all-MiniLM-L6-v2 sentence transformer model. This may take a few minutes depending on your internet connection.

Data Storage

Memories are stored locally in a ./memory_data directory relative to where you run the command. Each project should ideally run Super-Memory from its own directory to keep project-specific memories separate.

License

MIT License - see LICENSE