Skip to content

Hjarni vs Joplin

Joplin is open-source, encrypted, and now has an official MCP server. Hjarni's MCP server is hosted, so assistants reach it without your laptop.

Free to start. No credit card required.

Quick answer

Joplin is an open-source, end-to-end encrypted notes app that added an official MCP server in version 3.7. That server is local: it runs on the Web Clipper service at 127.0.0.1, so it works with desktop clients on the same machine and cannot be reached by ChatGPT on the web or Claude on your phone. Hjarni is a hosted Markdown knowledge base whose MCP server is a remote HTTPS endpoint any client can authorize over OAuth. Pick Joplin if you want open source, encryption you hold the keys to, and a desktop-only AI setup. Pick Hjarni if you want the same notes readable and writable from every assistant and every device.

Hjarni Joplin
AI access to notes

Both ship an official MCP server. The difference is where it listens, not whether it exists.

Built in, hosted Built in, local (3.7+)
Reachable by ChatGPT on the web

Joplin's server is a local endpoint on 127.0.0.1 behind the Web Clipper. ChatGPT only accepts a remote HTTPS connector, so it cannot reach localhost.

Reachable from your phone

Hjarni is a hosted endpoint any MCP client can authorize. Joplin's is bound to the machine running the desktop app.

Needs an app running to serve AI

Joplin's MCP server rides the Web Clipper service, so both have to be on and the machine awake.

Desktop app + Web Clipper
Claude Desktop setup

Stdio is not implemented in Joplin's v1 server, so desktop clients are pointed through an mcp-remote shim at the local port and token.

OAuth, one click npx mcp-remote bridge
Per-tool write controls

Joplin's write tools are off until you enable them individually, which is a genuinely careful default.

Folder-level AI instructions Per-tool toggles
Markdown notes
End-to-end encryption

Joplin can encrypt notes so its own sync target cannot read them. Hjarni encrypts at rest in the EU but holds the keys.

Open source

Joplin is open-source and self-syncable. A real strength if that matters to you.

Choose your own sync backend

Joplin syncs to Joplin Cloud, Dropbox, OneDrive, S3, WebDAV, or a local folder.

Wiki-links and backlinks

Joplin links notes by id. Backlinks come from community plugins rather than the app itself.

Built-in Plugins
Folder-level AI instructions

Hjarni lets you set AI behavior per folder, so a research folder and a journal behave differently.

Built-in
Team collaboration

Joplin shares notebooks through Joplin Cloud. Multi-user knowledge bases are not its primary shape.

Built-in Sync only
Email notes in
Built-in (Pro) Joplin Cloud (paid)
Free tier

Two official MCP servers, two different reaches

This is a closer comparison than most on this site. Joplin 3.7, released in August 2026, shipped a real MCP server built into the app, alongside semantic search and a set of AI tools. It is official, it is documented, and it is not a community plugin you have to keep alive yourself. On the axis most of these pages turn on, whether AI access is first-class, Joplin and Hjarni agree.

Where they part is the address. Joplin's MCP server rides on the existing Web Clipper service and listens on 127.0.0.1. Hjarni's is a hosted HTTPS endpoint that any MCP client authorizes over OAuth. Everything below follows from that one difference.

What localhost costs you

ChatGPT only accepts remote connectors, an HTTPS URL it can reach from OpenAI's servers. It has no way to reach into your laptop. So the official Joplin MCP server, as shipped, is not something you can add to ChatGPT. The same applies to Claude on the web, Claude on iOS and Android, and any assistant that is not a process running beside Joplin itself.

Even for desktop clients there is a shim. Stdio is not implemented in Joplin's first server version, and Claude Desktop speaks stdio, so the documented path is to point it through npx mcp-remote at the local port with the Web Clipper token. It works, and it is one more moving part that has to keep working.

And the server only answers while Joplin is open with Web Clipper started. Close the laptop and your AI's memory closes with it. Hjarni's notes are on a server, so an assistant reads them at 3am from a phone with no machine of yours involved.

Joplin gives your desktop AI your notes. Hjarni gives every AI your notes.

What Joplin does better

Two things, and they are not small. Joplin is open-source, so nothing about its future depends on a company staying solvent, and you can read the code that touches your notes. And it offers real end-to-end encryption: you can sync through Joplin Cloud, Dropbox, or S3 while holding keys the sync target does not have. Hjarni encrypts notes at rest in the EU, but Hjarni holds those keys. If your threat model includes the vendor, Joplin wins that argument outright and no amount of MCP convenience changes it.

Joplin's write-tool defaults are careful, too. Each tool is toggled individually and the write ones stay off until you turn them on. Their own documentation is blunt that Joplin will not ask you to confirm each write once they are enabled, which is a more honest warning than most apps give.

A concrete workflow difference

You are on a train, on your phone, and you want Claude to pull up what you decided about a supplier three months ago and add today's follow-up. With Joplin, the notes are on a laptop at home that is asleep, so there is nothing to ask. With Hjarni, Claude searches the folder, reads the note, and writes the follow-up back as a visible revision you can review later.

Flip it around. You are at your desk, you do not want anything leaving the machine, and you want an assistant that can read your journal without a vendor ever seeing it. That is Joplin's day, and Hjarni cannot offer it.

Can you host Joplin's MCP server yourself?

Partly. Because the endpoint is HTTP rather than stdio, people have put it behind a tunnel or rebuilt it as a hosted worker so remote clients can reach it, and the Data API has supported third-party servers for a while. If you enjoy running that, it is a legitimate answer and it gets you most of the reach.

It is worth being clear about what you are signing up for: a public endpoint in front of your notes, an auth story you own, and an uptime expectation, so that a chat on your phone works. Hjarni's position is not that this is impossible. It is that most people do not want a second job.

When Joplin is the better fit

Pick Joplin if open-source matters, if end-to-end encryption is a requirement rather than a preference, if you want to choose your own sync backend, or if your AI use is a desktop client on one machine. It is free, it is mature, it is genuinely good software, and its MCP server is the real thing. None of that is faint praise.

Migration and practical questions

Joplin exports as Markdown, with notebooks as folders. Hjarni's Markdown ZIP importer preserves folders, wiki-links, and frontmatter, so a Joplin export lands more or less intact. The real question is not the file format. It is whether you want your knowledge to live somewhere your assistants can always reach, or somewhere only you can.

When to use Joplin

  • You want open-source and end-to-end encryption
  • You want to pick your own sync backend
  • Your AI use is a desktop client on one machine

When to use Hjarni

  • You want ChatGPT and Claude on the web to reach your notes
  • You want your knowledge available when your laptop is not
  • You want folder-level AI instructions and team sharing

Joplin's server waits for your laptop. Hjarni's is already awake.

Common questions

Common questions

Does Joplin have an MCP server?

Yes. Joplin 3.7 (August 2026) shipped an official MCP server built into the desktop app, so it is not a community plugin. It runs on the existing Web Clipper service, which means it listens on 127.0.0.1 and only answers while Joplin is open with Web Clipper started.

Can ChatGPT connect to Joplin?

Not to the official server as shipped. ChatGPT only accepts remote connectors, an HTTPS URL it can reach from OpenAI's servers, and Joplin's endpoint is bound to localhost. People have tunnelled or re-hosted it to get around this, but that is a self-hosting project you run and secure yourself. Hjarni's MCP server is hosted, so ChatGPT connects over OAuth in about two minutes.

Can Claude Desktop connect to Joplin?

Yes, with a shim. Stdio is not implemented in Joplin's first server version and Claude Desktop speaks stdio, so the documented path points it through npx mcp-remote at the local port and Web Clipper token. It works on the machine running Joplin, and only there.

Is Joplin more private than Hjarni?

On the axis that usually matters most, yes. Joplin is open-source and offers real end-to-end encryption, so you can sync through Joplin Cloud, Dropbox, or S3 while holding keys the sync target does not have. Hjarni encrypts notes at rest in the EU and does not train on your data, but Hjarni holds the keys. If your threat model includes the vendor, Joplin is the better answer.

Can I import my Joplin notes into Hjarni?

Yes. Export Joplin as Markdown, where notebooks come out as folders, and drop the ZIP into Hjarni's importer. Folders, wiki-links, and frontmatter are preserved.

Why pick Hjarni if Joplin's MCP server is official too?

Reach, not whether the server exists. Joplin's is local, so your notes are readable by an assistant only while your machine is awake and only by clients running beside it. Hjarni's is a hosted endpoint, so the same notes are readable and writable by ChatGPT and Claude on the web, on a phone, and by Cursor or Claude Code, without a laptop in the loop. Joplin also has no folder-level AI instructions, and shared notebooks through Joplin Cloud are not the same as a team knowledge base with its own AI rules.

Start here

Write once. You both remember.

Free to start. No credit card required.

5 out of 5 from 9 reviews 4,600+ notes a day read or updated by AI

Works with Claude and ChatGPT today.