Quickstart

We’ll connect to Engrammic, store an observation, and query it back.

Prerequisites

• An MCP client (Claude Code, Cursor, Windsurf)
• Beta access key

1. Configure MCP

Add Engrammic to your client config. For Claude Code, edit ~/.claude/settings.json:

{
  "mcpServers": {
    "engrammic": {
      "url": "https://beta.engrammic.ai/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

2. Test the connection

Ask your agent:

“Use Engrammic to recall anything”

You should see a response, even if empty. That means it’s connected.

3. Store an observation

“Remember that I prefer TypeScript over JavaScript”

This calls:

{
  "tool": "remember",
  "arguments": {
    "observation": "User prefers TypeScript over JavaScript"
  }
}

4. Query it back

“What do you know about my language preferences?”

{
  "tool": "recall",
  "arguments": {
    "query": "language preferences"
  }
}

You’ll get your observation back with metadata about when it was stored and its decay class.

5. Store a fact

Facts need sources. Ask your agent:

“Learn that our API rate limit is 100 req/min, based on https://docs.example.com/limits”

{
  "tool": "learn",
  "arguments": {
    "claim": "API rate limit is 100 req/min",
    "evidence": "https://docs.example.com/limits"
  }
}

Next

• Working with Memory for patterns
• Agent Usage for decision guidelines
• MCP Tools for the full reference