This page covers connecting Claude Desktop to a remote brain. For a brain on the same machine as Claude Desktop, a local stdio entry in claude_desktop_config.json with "command": "gbrain", "args": ["serve"] works too — but only against a full local install, never a thin-client one.

For remote setup, first select native OAuth/PKCE or a private machine handoff according to the connection settings available in your installed Claude product. To open the owner dashboard or manage its clients, use MCP administration; the harness's OAuth scope does not grant that authority.

Important: Claude Desktop does NOT connect to remote MCP servers via claude_desktop_config.json. That file only works for local stdio servers. Remote HTTP servers must be added through the GUI.

Say to your agent: "connect claude desktop to my brain" — on the brain host, the remote-mcp skill publishes with gbrain mcp expose when needed, then MCP administration supplies the URL and setup for your chosen authentication method. You finish in the Claude Desktop GUI below.

Claude Desktop runs on your own device, so the tailnet-only default is enough — nothing is exposed to the public internet:

gbrain mcp expose   # prints https://your-machine.your-tailnet.ts.net/mcp

If the brain already has an HTTPS endpoint, use it instead. Register a native OAuth client or provision a private machine handoff through the running server's owner API, according to the installed client's settings. The expose-managed owner credential is at ~/.gbrain/serve/admin-token; it is never a client credential.

Say to your agent: "put my brain on tailscale" — "connect claude desktop to my brain".

Legacy bearer settings on PGLite: gbrain auth create "claude-desktop" --scopes read,write opens the database, which fails with live_serve while the expose-managed service holds it. Mint the token before the service runs (ahead of gbrain mcp expose, or while the service is stopped briefly), or provision through the running server instead — gbrain mcp grant … --admin-token-file ~/.gbrain/serve/admin-token or the /admin dashboard. Postgres brains mint fine while the server runs.

The device running Claude Desktop must be on the same tailnet (Tailscale installed and signed in). Full walkthrough and troubleshooting: remote MCP guide. Using ngrok instead? Its URL is https://YOUR-DOMAIN.ngrok.app/mcp (ngrok-tunnel recipe).

1. Open Claude Desktop
2. Go to Settings > Integrations
3. Click Add Integration (or Add Connector)
4. Enter the MCP server URL:

   https://your-machine.your-tailnet.ts.net/mcp
   

   Use your MagicDNS name as printed by gbrain mcp expose, or the existing HTTPS endpoint (ngrok alternative: https://YOUR-DOMAIN.ngrok.app/mcp).
5. Choose the authentication method that the settings support. For native OAuth, use the owner-issued client metadata and the native connection procedure. For an existing bearer connection, enter its private scoped token. Never enter the server's owner bootstrap credential as the MCP credential.
6. Save

Start a new conversation and try:

Search my brain for [any topic]

Observe the actual GBrain tool call and result. A saved configuration alone does not establish that this Claude Desktop session loaded or connected the server.

Using claude_desktop_config.json for remote servers — this silently fails with no error message. The JSON config only works for local stdio MCP servers. Remote HTTP servers must be added via Settings > Integrations in the GUI.

Using the wrong URL — make sure the URL ends with /mcp (not /health or just the base domain).

Name does not resolve on this device — the device is not on the tailnet, or MagicDNS is off for it. Sign in to Tailscale on the device and enable "Use Tailscale DNS settings"; gbrain mcp expose --status on the host confirms the server side. See the troubleshooting table.