How do I connect ChatGPT or Claude to Harbor?

You connect your AI through Harbor’s open surface — the CLI, the agent skill, and the MCP server.

The quickest path, if you use an AI coding assistant:

1. Create a Personal Access Token in Harbor. This is how your AI signs in as you — and you can revoke it any time.
2. Install the harbor CLI.
3. Run harbor skill install. It drops a self-contained agent skill into Claude Code, Codex, or Cursor that teaches the assistant to create, edit, richly format, organize, search, remind, and share notes.

From there, you just ask: “file these meeting notes into the Projects notebook and tag them follow-up,” and your assistant does it through the CLI.

Using Claude or ChatGPT to chat rather than to code? Skip the CLI. Add Harbor as a custom connector instead — paste https://app.harbor.my/mcp, approve the login, and your notes become a tool the AI can use directly. Nothing else to install: here are the steps.

More on how the two compare in What are the Harbor CLI and MCP server?, and for the wider picture — who else ships one — MCP servers for notes, explained.

To point an assistant at one particular note rather than describing it and hoping search finds the right one, paste in the note’s ID — one click copies it from the note’s Info panel.

Building something yourself? Every CLI command supports --json, and harbor openapi hands any tool the complete OpenAPI spec. Third-party apps can integrate via OAuth.

Two privacy notes, plainly stated. Your AI sees only what you’ve authorized it to see, and over the API or MCP an encrypted note isn’t part of that — here’s exactly what it gets instead. The CLI is the one exception, because it runs on your machine: set HARBOR_PASSPHRASE in a terminal an agent can reach and it can read your encrypted notes there. That’s your call to make, not a leak — but make it knowingly. And what your AI provider does with text you send it is governed by their policies, so connect the AI you trust.