r3 Documentation

MCP server for persistent AI memory. Local Redis storage with optional cloud sync.

Start building

How r3 works

r3 is an MCP server that gives AI applications persistent memory. By default, it uses an embedded Redis server for local storage. No setup required. Optionally add cloud sync with Mem0.

json
1
2
3
4
5
6
7
8
9
// MCP client config (e.g. .gemini/settings.json)
{
  "mcpServers": {
    "r3": {
      "command": "npx",
      "args": ["@n3wth/r3"]
    }
  }
}

Once configured, your AI assistant can remember context across conversations:

text
1
2
3
4
5
6
You: Remember that I prefer TypeScript over JavaScript
AI: I'll remember your preference for TypeScript.
​
[New conversation]
You: What language should I use for this project?
AI: Based on your preferences, I'd recommend TypeScript.

Architecture

Core concepts

Memories

Each memory contains:

• Content: The information to remember
• User ID: Who it belongs to
• Priority: How important it is (affects caching)
• Metadata: Additional context

Caching layers

1. L1 (Redis): Hot data, 24-hour TTL
2. L2 (Cloud): Permanent storage with vector search

MCP tools

r3 provides these tools to MCP clients:

• add_memory: Store new information
• search_memory: Find relevant context
• get_all_memories: List all memories
• delete_memory: Remove specific memory
• cache_stats: Monitor performance

When to use r3

Good for:

• User preferences and settings
• Conversation history
• Facts about users or projects
• Context that should persist

Not for:

• Sensitive data (use encryption)
• Large files (use object storage)
• Real-time data (use streaming)

Get help

• GitHub Issues