@openengram/mcp

MCP (Model Context Protocol) server for Engram — persistent memory for AI agents. Works with Claude Desktop, Cursor, and any MCP client.

Features

• 6 Tools: engram_remember, engram_recall, engram_search, engram_forget, engram_context, engram_observe
• 2 Resources: engram://stats, engram://context
• 1 Prompt: memory-aware-chat
• Security: Input validation, rate limiting, TLS enforcement for remote backends
• Offline-resilient: Graceful degradation when Engram is unreachable
• Zero disk footprint: Stateless proxy — stores nothing locally

Quick Start

1. Install

npm install -g @openengram/mcp

2. Configure Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

Cloud (works anywhere — recommended):

{
  "mcpServers": {
    "engram": {
      "command": "engram-mcp",
      "env": {
        "ENGRAM_API_URL": "https://api.openengram.ai",
        "ENGRAM_API_KEY": "your_api_key",
        "ENGRAM_USER_ID": "your_user_id"
      }
    }
  }
}

Self-hosted (local network):

{
  "mcpServers": {
    "engram": {
      "command": "engram-mcp",
      "env": {
        "ENGRAM_API_URL": "http://localhost:3001",
        "ENGRAM_API_KEY": "your_api_key",
        "ENGRAM_USER_ID": "your_user_id"
      }
    }
  }
}

3. Restart Claude Desktop

Quit and reopen Claude Desktop. You should see the Engram tools available in the chat input.

Environment Variables

| Variable | Required | Default | Description | |----------|----------|---------|-------------| | ENGRAM_API_KEY | ✅ | — | Your Engram API key (eng_ prefix) | | ENGRAM_USER_ID | ❌ | Beaux | User namespace ID | | ENGRAM_API_URL | ❌ | https://api.openengram.ai | Engram API URL | | ENGRAM_BASE_URL | ❌ | — | Alias for ENGRAM_API_URL | | ENGRAM_TIMEOUT_MS | ❌ | 10000 | Request timeout (ms) | | ENGRAM_LOG_LEVEL | ❌ | warn | debug / info / warn / error | | ENGRAM_MAX_RETRIES | ❌ | 2 | Retry count for transient failures | | ENGRAM_ALLOW_HTTP | ❌ | false | Allow non-localhost HTTP (dev only) |

Tools

| Tool | Description | |------|-------------| | engram_remember | Store a memory with optional layer, importance, and tags | | engram_recall | Semantic search across memories with relevance scores | | engram_search | Entity/graph-aware search | | engram_forget | Delete a memory by ID | | engram_context | Generate an LLM-optimized context window | | engram_observe | Auto-extract and store memories from text |

Development

git clone https://github.com/heybeaux/engram-mcp.git
cd engram-mcp
npm install
npm run build

# Run locally
ENGRAM_API_KEY=your_key ENGRAM_USER_ID=dev npm run dev

Links

• Engram — The memory backend
• OpenEngram — Hosted cloud service
• Dashboard — Web UI for managing memories

License

MIT