Set up Hjarni by talking to your LLM

Hjarni publishes a single doc URL that any LLM can read. Paste it into ChatGPT, Claude, Cursor, or Claude Code and your AI knows how to connect the MCP server, propose a folder structure, and seed your first notes. You never leave the chat.

The two URLs you need

Everything else on this page is detail. If you only remember two things, remember these.

Paste in your LLM https://hjarni.com/llms-full.txt

MCP server URL https://hjarni.com/mcp

The first is a plain-text document containing the MCP endpoint, the OAuth metadata, every tool the server exposes, and a step-by-step onboarding script. Any LLM that can fetch a URL can read it and act on it.

Starter prompt

Paste this into ChatGPT, Claude, Cursor, or Claude Code after you have signed up for Hjarni:

“Please read https://hjarni.com/llms-full.txt and help me set up Hjarni. Confirm the MCP connection works, then propose a starter folder hierarchy and seed my first notes based on what I tell you I want to capture.”

The AI fetches the doc, walks you through the connection step for your specific client, and then proposes a starter folder hierarchy you can accept or refine before any notes are created.

Per-client install snippets

Sometimes the LLM cannot complete the connection itself (Claude.ai needs you to click through a Connectors menu; Cursor reads a config file). These are the one-line snippets to use for each client.

Claude.ai (web and iOS)

Open Customize > Connectors, click Add custom connector, paste the MCP URL, and complete the OAuth flow.

See the full Claude setup guide for per-plan instructions.

Claude Desktop

Same as Claude.ai for remote servers. Open Customize > Connectors, paste the MCP URL, complete OAuth.

Claude Code

From the command line:

claude mcp add --transport http --scope user hjarni https://hjarni.com/mcp

--scope user makes the server available across every project. Claude Code handles OAuth in your browser.

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "hjarni": {
      "url": "https://hjarni.com/mcp"
    }
  }
}

ChatGPT

In a Pro/Team/Enterprise account, open a chat, click the +, add a Custom MCP connector, and paste the MCP URL.

See the full ChatGPT setup guide for the per-plan flow and the dev-mode toggle.

Raw curl (smoke test)

If your LLM client cannot do MCP, every tool also has a REST endpoint. Generate an API token in your account settings and hit:

curl -H "Authorization: Bearer $HJARNI_TOKEN" https://hjarni.com/api/v1/dashboard

See the REST API reference.

What happens after the connection

Once your LLM can reach the MCP server, the onboarding script in llms-full.txt tells it to:

Read any account-wide AI instructions you have already set.
Check the dashboard. If your account is empty, treat this as a fresh onboarding.
Ask which persona fits best (developer, founder, researcher, writer, or traveler) and what you want to capture.
Propose a starter folder hierarchy and wait for your approval before creating anything.
Seed a small set of useful starter notes in the folders, with summaries and tags already filled in.
Offer to set folder-level AI instructions so future sessions follow the same conventions automatically.

From that point your LLM is reading and writing notes through the same MCP server every time you start a new conversation.

Troubleshooting

The LLM says it cannot fetch the URL

Some chat clients block outbound HTTP unless you explicitly enable a "web" or "browse" tool. Turn that on, or paste the contents of llms-full.txt directly into the chat.

The LLM connected MCP but tool calls return 401

The OAuth flow probably did not complete. Disconnect the server in your client's connectors menu and add it again. The browser handoff should land on a Hjarni consent screen.

The LLM creates folders without asking

Repeat the starter prompt and add: "Show me your plan before creating anything." The doc instructs the LLM to confirm before mutating, but some clients are more eager than others.

The LLM keeps offering tools that look wrong

It is probably using a cached older copy of the doc. Ask it to refetch https://hjarni.com/llms-full.txt. The file is auto-generated from the live tool definitions, so it is always current.