# Capture into Hjarni from the Alfred launcher

Alfred is the keyboard-first launcher. Hjarni is the knowledge base. One workflow connects them, no AI required.

25 notes free, forever. No credit card required.

## Connect Alfred in about two minutes

1. Sign up for Hjarni and generate a personal API token at hjarni.com/settings/connections.
2. Store the token in an Alfred workflow variable HJARNI_TOKEN. Mark it as secret so it stays out of exports.
3. Create a new Alfred workflow with a Keyword input (for example 'hj') feeding into a Run Script bash action.
4. Paste the curl. Use $1 for the title and $2 for the body, or split into two steps with a Large Type confirmation in between.
5. Trigger Alfred, type your keyword, and confirm the note shows up in Hjarni. Add a Script Filter for search if you want recall too.

```
# Alfred workflow → Run Script (bash):
curl -s -X POST https://hjarni.com/api/v1/notes \
  -H "Authorization: Bearer $HJARNI_TOKEN" \
  -H "Content-Type: application/json" \
  -d "{\"title\": \"$1\", \"body\": \"$2\"}"
```

The Alfred workflow calls Hjarni's REST API at `https://hjarni.com/api/v1/` and authenticates with a bearer token. Generate one at hjarni.com/settings/connections and store it in the workflow as a secret variable.

## Workflows that actually use your context

1. Capture a note in three keystrokes

   Trigger Alfred, type your keyword, type the note. The workflow posts it to Hjarni's REST API and the note lands in your library on the next page load.

2. Open the right note straight from the launcher

   Type a query in Alfred and pick from the matching Hjarni notes. Hit enter and Hjarni opens at the right note in your browser.

3. Pair capture-from-Alfred with read-elsewhere

   Alfred handles fast keyboard capture on the desktop. If you also use an MCP client like Claude or Cursor, they read the same Markdown notes: one source of truth, two ways in.

4. Keep the launcher AI-free if you want

   Alfred has no built-in AI and that is fine. The workflow uses Hjarni's REST API directly. You get capture and recall without sending anything through a model.

5. Power-user keywords per folder

   Wire one keyword per folder if you have favorite spots: 'meet' goes to Meetings, 'idea' to Ideas. The workflow is plain JSON, easy to fork and extend.

## Why Hjarni works with Alfred without MCP

Alfred is the rare modern launcher that ships without an AI assistant. That is by design and it is part of why people love it. The right way to bring Hjarni into Alfred is the same way Alfred talks to every other service: a workflow that hits a REST API.

Hjarni's REST API exists on every plan, including the free tier, and uses bearer tokens. The same endpoints that power the MCP server power the API. Notes you create from Alfred are the same notes Claude reads via MCP in the browser.

If you already use Alfred for clipboard, snippets, and app launching, this is the natural place for a capture-to-Hjarni workflow. It costs you one keyword and zero tabs, and the notes you capture are visible to whichever MCP-aware assistants you also connect to your account.

Common questions

## Questions before you connect Alfred

**Does Alfred support MCP?**

No. Alfred has no built-in AI and no MCP client. This integration uses Hjarni's REST API from an Alfred workflow, which is how Alfred talks to every other service.

**Do I need Alfred Powerpack?**

Yes. The Run Script action used in the workflow requires Powerpack. The keyword input and Large Type are also Powerpack features.

**How do I store the Hjarni token safely in Alfred?**

Use a workflow variable and set it to secret. Secret variables stay out of workflow exports, so you can share the workflow with someone else without leaking your token.

**Can I search Hjarni from Alfred, not just create notes?**

Yes. Add a Script Filter that calls GET /api/v1/search and emits the results as Alfred items. Enter on a result can open the note URL or paste a quoted summary.

**Is this slower than using an MCP client?**

It is a single HTTP request, so the round trip is fast. The difference is shape: Alfred is for keystroke-level capture and recall, an MCP client is for conversational reasoning. Pair the two and you have both.