Engram

Engram MCP lets your AI agents remember past conversations, facts, and decisions by connecting to a hosted memory service. It makes interactions feel more continuous and gives you tools to store, search, and manage memories across conversations.

How to use

Connect your MCP client to Engram to start storing and recalling memories. You can store individual memories or collect multiple memories each turn, search for relevant past items, browse all memories, delete specific memories, and explain why a retrieved result was chosen. Use memory to preserve user preferences, recurring tasks, decisions, and outcomes, then ground responses in those memories for improved continuity and personalization.

How to install

Prerequisites: You only need access to Engram’s hosted MCP endpoint and a valid API key. No local server installation is required.

1. Get your API key. Sign up at lumetra.io to obtain your API key.

2. Add Engram to your MCP client. Choose one of the following client configurations.

claude mcp add-json engram '{"type":"http","url":"https://engram.lumetra.io","headers":{"X-API-Key":"<your-api-key>"}}'

Additional configuration details

Engram is accessible as a remote MCP server at the URL https://engram.lumetra.io. You can provide the API key via the X-API-Key header for authorization. You may use the header form in MCP client configurations as shown in the example commands.

{
  "mcpServers": {
    "engram": {
      "serverUrl": "https://engram.lumetra.io",
      "headers": {
        "X-API-Key": "<your-api-key>"
      }
    }
  }
}

Security and maintenance notes

Keep your API key secret. Rotate keys if you suspect exposure. The Engram service includes memory controls for retention and cleanup, so you can remove outdated or incorrect memories and re-store corrected facts.

Available tools

store_memory

Store a single memory to Engram so it can be retrieved later

store_memories

Store multiple memories in a batch for efficient ingestion

search_memories

Search stored memories with a query to retrieve relevant results

memory_index

Browse all stored memories, with optional pagination

delete_by_event

Delete memories by their event id, with an option for a dry run

explain_retrieval

Explain why a memory retrieval was chosen for a given result