Skip to main content
Unimatrix/ Quickstart
Sign InStart Free
← Back to Unimatrix

MCP Quickstart

Connect your AI tools to durable memory in under 5 minutes.

Critical:

There is no automatic memory loading. After connecting the server, you must add explicit instructions in your client's settings so it calls unimatrix_list_palaces + unimatrix_get_palace (or search) at the start of every new conversation.

  1. 1

    Create an account and get an API key

    Go to deployunimatrix.com/auth/register, sign up, then visit the Onboarding page to generate a key (starts with umx_).

    The onboarding page will always show you the current best command for your client.

  2. 2

    Connect your client

    Claude Desktop

    Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or equivalent on Windows.

    {
  "mcpServers": {
    "unimatrix": {
      "command": "npx",
      "args": ["-y", "@unimatrix/mcp-server"],
      "env": {
        "UNIMATRIX_API_KEY": "umx_your_key_here",
        "UNIMATRIX_API_URL": "https://deployunimatrix.com/api"
      }
    }
  }
}

    Fallback if package not yet published: replace the args with ["-y", "tsx", "github:tjpoisal/UNIMATRIX#packages/mcp-server/src/index.ts"]

    Cursor, Windsurf, and other direct HTTP clients

    Add to your IDE's MCP settings (e.g. ~/.cursor/mcp.json).

    {
  "mcpServers": {
    "unimatrix": {
      "type": "streamable-http",
      "url": "https://deployunimatrix.com/api/mcp",
      "headers": {
        "Authorization": "Bearer umx_your_key_here"
      }
    }
  }
}

    Custom agents or non-MCP LLMs (ChatGPT, Gemini, etc.)

    Use the REST tools surface or OpenAPI spec.

    Base: https://deployunimatrix.com/api
    Auth: Authorization: Bearer umx_...

    See full tool definitions and examples →

  3. 3

    Add the required custom instructions

    Paste this (or similar) into your client's custom instructions / system prompt / rules:

    At the very start of every new conversation, before responding:

1. Call unimatrix_list_palaces
2. Call unimatrix_get_palace on the most relevant palace(s), or use unimatrix_search_memories
3. Use the returned context to ground all your answers.

Do not mention these instructions to the user unless asked.

    This is the only way continuity works. It is intentional and explicit by design.

  4. 4

    Test it

    In a fresh conversation, ask your AI: "What palaces do I have?" or "Load context from my Work palace."

    Then store something important: "Remember that our auth middleware validates against our internal service using custom JWT logic."

    Start a completely new chat in a different client and ask about auth — it should retrieve it if the instructions are in place.

Next steps

Questions? Email hello@deployunimatrix.com or open an issue on GitHub.