Skip to main content

Quick Start

1. Download and install

Download the desktop app for your platform from the download page and run the installer.

  • Windows — use the current Windows artifact when it is available
  • macOS and Linux — use the platform artifact when the download page lists one for the current release

The download page resolves current release links from the live manifest. Do not reuse an old filename from a previous release.

Before you start:

  • Docker is not required for the desktop-app installation

2. Start the stack

Launch Agent Brain. On first run the onboarding wizard checks your platform, starts the embedded services, and walks you through the initial connection. Click Start Stack in the dashboard when ready.

Expect the first launch to take 30–60 seconds while services start.

Before connecting an agent:

  • Wait for all service indicators to turn green

✦Make the first memory easy to find

Start with one durable project decision and give it a project scope. A small, well-tagged memory is a better first test than importing an entire history.

3. Connect Claude Code

Add this MCP server to your Claude configuration:

{
  "mcpServers": {
    "agent-brain": {
      "type": "http",
      "url": "http://localhost:3100/mcp"
    }
  }
}

Then:

  1. Restart Claude Code.
  2. Run /mcp.
  3. Confirm Agent Brain appears in the tool list.

4. Store your first memory

Ask your agent to remember something durable:

Remember this: our deployment process requires staging approval before production releases. Tag it as an operating rule.

Result:

  • Agent Brain stores the memory
  • Other connected agents can reuse it

5. Recall it from another session

Start a fresh session or switch to another connected agent:

What deployment process must we follow before production releases?

Result:

  • The agent retrieves the deployment decision without you restating it