r3

npm version npm downloads License: MIT

Persistent memory for MCP clients (Claude Desktop, Claude Code, Cursor) that runs entirely on your machine, with no accounts, no API keys, and no cloud service required to start.

Memory persisting across two separate server processes

Quick start

npx @n3wth/r3

That single command starts an MCP server backed by an embedded Redis instance and a local vector index. There is no separate database to install and no signup step.

How it works

MCP client (Claude Desktop / Claude Code / Cursor)
        |
        v
   r3 MCP server (stdio)
        |
   +----+-----------------------+
   |                            |
embedded Redis            vectra local index
(redis-memory-server,      (on-disk vector store
 auto-downloaded binary,    for semantic search,
 no external service)       no external service)
   |
   +--- optional ---> Mem0 cloud API (MEM0_API_KEY)
                       cross-device sync, off by default

Embedded Redis — redis-memory-server downloads and manages a local Redis binary for you. It stores memory content and metadata. If it cannot start, r3 falls back to an in-process store.
vectra — a local, file-backed vector index used for semantic search. No network calls, no external vector database.
Mem0 (optional) — if MEM0_API_KEY is set, r3 also syncs to Mem0's cloud API so memories can follow you across machines. Without a key, nothing leaves your machine.

MCP client configuration

Claude Desktop

Edit claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "r3": {
      "command": "npx",
      "args": ["@n3wth/r3"]
    }
  }
}

Claude Code

claude mcp add r3 "npx @n3wth/r3"

Cursor

Add to .cursor/mcp.json in your project (or the global Cursor MCP config):

{
  "mcpServers": {
    "r3": {
      "command": "npx",
      "args": ["@n3wth/r3"]
    }
  }
}

Restart the client after editing config. In a new conversation:

You: Remember that I prefer TypeScript and dark mode.
AI: I'll remember that.

[new conversation]

You: What are my preferences?
AI: You prefer TypeScript and dark mode.

Tools

Tool	Description
`add_memory`	Store content with optional metadata and priority
`search_memory`	Query memories using semantic or keyword search
`get_all_memories`	List stored memories with pagination
`get_memory`	Retrieve a specific memory by ID
`update_memory`	Modify existing memory content or metadata
`delete_memory`	Remove a memory
`deduplicate_memories`	Find and merge duplicate memories
`cache_stats`	Report cache hit rate and storage stats
`sync_status`	Report Mem0 cloud sync status
`optimize_cache`	Run cache maintenance
`import_memories`	Bulk import memories

Enhanced mode (default, INTELLIGENCE_MODE=enhanced) adds:

Tool	Description
`extract_entities`	Extract named entities from text
`get_knowledge_graph`	Return the entity/relationship graph
`find_connections`	Find entities connected to a given entity

Configuration

Environment variables, all optional:

Variable	Description	Default
`REDIS_URL`	Use an external Redis instead of the embedded one	embedded server
`MEM0_API_KEY`	Enables Mem0 cloud sync	unset (local only)
`MEM0_USER_ID`	Namespace for memories	`default`
`INTELLIGENCE_MODE`	`enhanced` or `basic`	`enhanced`

Example with cloud sync enabled:

{
  "mcpServers": {
    "r3": {
      "command": "npx",
      "args": ["@n3wth/r3"],
      "env": {
        "MEM0_API_KEY": "mem0_..."
      }
    }
  }
}

Comparison

	r3	mem0 (OSS)	zep
Runs fully local with zero config	yes (embedded Redis + vectra)	requires a Postgres/vector DB you configure	requires a Postgres instance you configure
Needs an API key to try it	no	no (self-hosted) / yes (cloud)	yes (cloud), or self-hosted setup
Optional cloud sync	yes, via Mem0	n/a (is the cloud option)	yes

This table only reflects setup requirements observed in each project's own documentation, not benchmark performance or feature completeness. Verify against current upstream docs before relying on it.

Known issues

See LAUNCH_AUDIT.md for current limitations, including a native module build failure on some macOS setups.

Documentation

Full documentation at r3.n3wth.com.

License

MIT

r3 (Recall)

r3