Start your first backpack.

1. Connect the MCP server

Install a current Node.js LTS release and use an MCP client that supports local stdio servers. Backpack's packages declare Node.js 18 or later; use a supported LTS release for a new installation.

Add this server to your client's MCP configuration. The location of that configuration depends on your client.

{
  "mcpServers": {
    "backpack": {
      "command": "npx",
      "args": ["--yes", "backpack-ontology@1.1.0"],
      "env": { "DO_NOT_TRACK": "1" }
    }
  }
}

This pins the current engine release and disables package telemetry. Restart or reconnect your client, then check that Backpack's tools are available. The first run downloads the package from npm.

2. Give it something to remember

Ask your AI client:

Create a learning graph called project-notes. Add a project named Demo Garden, record that it uses drip irrigation, and connect the two.

Then ask it to list your graphs and read back what it saved. Backpack provides the tools; your AI client decides when to call them.

3. Open the local viewer

npx --yes backpack-viewer@1.0.1

Open http://localhost:5173 and select your graph. Run the viewer under the same user and with the same Backpack directory as the MCP server.

The viewer is a local application. This website's interactive demo uses a bundled public example and does not access your local graphs.

Using the Claude Code plugin?

The Backpack plugin adds usage guidance for graph building and research. Choose either its MCP setup or the standalone configuration above to avoid duplicate servers. Configure DO_NOT_TRACK=1 in the environment launching your client if you use the plugin.

No Backpack account required

The local tools do not require a Backpack subscription. Your chosen AI client or model provider may charge separately. The former hosted service, cloud sign-in and cloud sync are no longer available.

Read about storage and backups or ask the community for help.