I Gave Claude Access to My Notes. Here's What Happened.

Every conversation with Claude starts the same way. You explain who you are. What you're working on. What you decided last week. You paste in context. You summarise things Claude should already know.

It's like having a brilliant colleague with amnesia.

I got tired of it. So I built a note-taking app that Claude can read. Not through copy-paste. Not through a plugin. Through MCP. Claude searches my notes, reads them, creates new ones, and links them together. Directly.

Here's what happened.

What's MCP?

MCP is an open protocol from Anthropic. It lets Claude call your apps directly. Search your notes. Read a file. Create a link between two ideas. Instead of pasting context into a chat window, Claude reaches into your app and pulls what it needs.

In my case, the app is Hjarni. Plain Markdown notes. Folders. Tags. Links between notes. Nothing fancy. The point isn't the editor. It's the connection.

Setup takes two minutes. You add the MCP server URL in Claude's settings. Done.

The moment it clicked

I was planning a campervan road trip through Sweden with my family. I'd been collecting notes for weeks. Campgrounds in Småland. Local festivals in Skåne. Loppisar (Swedish flea markets) I wanted to visit.

Mid-conversation, I asked Claude: "What festivals are happening in southern Sweden in August?"

Claude didn't ask me to paste anything. It searched my notes. Found six events I'd saved across different research sessions. Pulled them together. Suggested a route that hit three of them.

Not because the answer was brilliant. I could have done it myself with enough clicking. But Claude knew where to look. Like a real collaborator would.

What this looks like day-to-day

This isn't a hypothetical demo. I use Hjarni every day. Here's what it actually looks like:

Flat tyre, no panic. I got a flat on my gravel bike. Asked Claude which inner tube to buy. It pulled up my bike's spec sheet, found my Canyon Grizl 7 with Schwalbe G-One Bite 45mm on 700c wheels, and recommended the right tube. I didn't look up a single spec.

Research that compounds. Every article I read about Sweden, every campground I bookmark, every festival date: it goes into Hjarni. When I plan the next leg of the trip with Claude, all that research is there. Claude doesn't start from zero. It starts from everything I've collected.

People notes. I keep short notes about people. Where we met, what we talked about. Before a meeting, I ask Claude to pull up the note. Like a CRM that doesn't make you hate yourself.

Project context. I'm building Hjarni solo alongside a full-time job. Every product decision, feature idea, and pricing discussion lives in my notes. When I'm weighing a new feature, Claude pulls up what I've already considered and what I decided against. Including why.

Writing for two readers

I used to write notes for myself. Quick. Messy. Full of shorthand only I'd understand.

Now I have a second reader. And that reader is literal.

So I started writing differently. Not for AI. Just clearer. A real title instead of "stuff for Sweden trip." Tags that mean something. A one-line summary at the top.

People have been talking about "second brains" for years. The problem was never the editor. It was the payoff. A second brain you never open is just a graveyard.

AI changes that. Every note you write makes your next conversation better. That's the loop. And the side effect? Making your notes machine-readable makes them more human-readable. That wasn't the plan. But it's the best thing I've found.

This isn't "AI memory"

I should be clear about what this is not.

ChatGPT and Claude both have "memory" features. A handful of bullet points the AI stores between conversations. It's shallow. You can't see most of it. You can't organise it. You can't edit it meaningfully.

This is different. Your notes are yours. You see exactly what Claude can access. You organise it. You decide what stays. Claude doesn't silently build a profile of you. It reads what you've written, when you ask it to.

With AI memory, you're trusting a black box. With notes, you're trusting yourself.

Try it

Write your first note. Connect Claude. Ask it something only your notes can answer.

It's free to start. You'll know it's working when Claude stops asking you to explain yourself.