Local LTM (Long-Term Memory)

A modular, locally-maintained Long-Term Memory system for AI agents powered by Mem0 and exposed via MCP (Model Context Protocol).

Store, search, and manage persistent knowledge for AI agents across conversations.

Setup

1. Prerequisites

• Node.js 16+
• Mem0 API Key (get here)

2. Installation

npm install

3. Configuration

Create a .env file:

MEM0_API_KEY=your_mem0_api_key_here

See .env.example for more details.

Quick Start

Start Server

node src/index.js

Test

node src/test.js

Claude Desktop Integration

Add to your Claude Desktop config file:

Windows: %APPDATA%\Claude\claude_desktop_config.json
macOS/Linux: ~/.claude_desktop_config.json

{
  "mcpServers": {
    "long-term-memory": {
      "command": "node",
      "args": ["C:\\Users\\YourUsername\\Desktop\\local-ltm\\src\\index.js"],
      "env": {
        "MEM0_API_KEY": "your_mem0_api_key_here"
      }
    }
  }
}

Then restart Claude Desktop. The LTM tools will be available!

Available Tools

• memory.store - Save memories (semantic, episodic, procedural)
• memory.search - Query memories by namespace
• memory.update - Update existing memories
• memory.delete - Remove obsolete memories

Memory Organization

Shared Knowledge:

/shared/user_preferences
/shared/facts

Agent-Specific:

/agents/{agent_id}/private/heuristics

Memory Types

• episodic - Events, experiences, conversations
• semantic - Facts, knowledge, stable information
• procedural - Preferences, rules, best practices

Project Structure

src/
├── index.js           # MCP Server
├── mem0.js            # Mem0 API wrapper
├── test.js            # Test client
├── config/
│   ├── prompts.js
│   └── server.js
└── tools/
    ├── store.js
    ├── search.js
    ├── update.js
    └── delete.js

Usage Example

// Store a memory
await client.send("tools/call", {
  name: "memory.store",
  arguments: {
    namespace: "/shared",
    content: "User prefers concise explanations",
    type: "procedural",
    confidence: 0.9
  }
});

// Search memories
const results = await client.send("tools/call", {
  name: "memory.search",
  arguments: {
    namespace: "/shared",
    query: "user preferences"
  }
});

Guidelines

Store when:

• Information is durable and reusable
• Information is stable (preferences, facts, decisions)

Don't store:

• Temporary context
• Task-specific details
• Raw conversation text (unless critical)

Adding Tools

1. Create src/tools/mytool.js
2. Export tool with schema and execute function
3. Register in src/tools/index.js

License

MIT