Hjarni for Junie

The AI-native knowledge base for Junie

Junie is JetBrains' AI coding agent. Hjarni is the knowledge it reads while it works: your conventions, architecture notes, and runbooks, available in the IDE without re-explaining them.

Get started free See how it works

Free to start. No credit card required.

What this unlocks

Workflows that actually use your context

Conventions Junie can actually follow

Junie writes and edits code in your JetBrains IDE. Keep your conventions, patterns, and decisions in Hjarni and Junie reads them instead of guessing house style.

Context beyond the open project

Junie sees your code. Hjarni holds what the code does not say: why a decision was made, how a deploy runs, what failed before. Junie pulls that in on demand.

Write the decision back

Hjarni's MCP tools include create and update, so Junie can record an architectural decision or a fix as a note that the next session, and your teammates, read.

Shared memory across the team

The same Hjarni notes every developer's Junie reads are the ones their other AI clients read, so the whole team works from one source of truth.

Setup

Connect Junie in about two minutes

1
Sign up for Hjarni. Make sure Junie is enabled in your JetBrains IDE (it reached general availability in 2026).
2
Open Junie's MCP Installation Assistant and add a remote server, or edit .junie/mcp/mcp.json (project) or ~/.junie/mcp/mcp.json (user).
3
Use the URL https://hjarni.com/mcp. From the assistant, choose Authorize to complete the OAuth sign-in.
4
Prefer a token? Add an Authorization Bearer header with a Hjarni personal API token instead of authorizing via OAuth.
5
Ask Junie to list your Hjarni notes to confirm the connection is active.

Add Hjarni to Junie's mcp.json

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

Add the server with Junie's MCP Installation Assistant (then choose Authorize to sign in via OAuth in the browser), or edit Junie's mcp.json directly: project scope at .junie/mcp/mcp.json or user scope at ~/.junie/mcp/mcp.json. For a token instead of OAuth, add an Authorization Bearer header with a Hjarni personal API token. Check JetBrains' Junie MCP docs for the exact file shape.

Why a coding agent needs memory outside the repository

Junie is good at writing code once it understands the project. The gap is the context that never makes it into the repo: the reasoning behind a decision, the runbook for a release, the convention everyone follows but no one wrote down.

Hjarni is where that lives, as Markdown notes you own and export anytime. Junie reads the relevant notes while it codes, so it follows your patterns instead of inventing its own.

Junie supports remote MCP servers with OAuth or header auth, so Hjarni is a short mcp.json entry or a couple of clicks in the installation assistant. Scope it per project or per user, and the same notes serve your other AI clients too.

Common questions

Questions before you connect Junie

Does Junie support remote MCP servers?

Yes. Junie, JetBrains' AI coding agent, supports remote MCP, configured in its mcp.json or through the MCP Installation Assistant.

Where does Junie's MCP config live?

Project scope at .junie/mcp/mcp.json, or user scope at ~/.junie/mcp/mcp.json. Check JetBrains' Junie MCP docs for the exact file shape.

OAuth or token?

Both. Use the MCP Installation Assistant and choose Authorize for the OAuth flow, or add an Authorization Bearer header with a Hjarni personal API token in mcp.json.

Is this JetBrains AI Assistant or Junie?

This connects Junie, JetBrains' coding agent, which has documented remote MCP support. JetBrains AI Assistant's own MCP support is on a separate track.

What should I keep in Hjarni for Junie?

Team conventions, architecture decisions, runbooks, and the reasoning behind past changes: the context that is not in the repository but that Junie needs to follow your patterns.

Give Junie a memory

The session starts where the last one ended. Write notes once. Junie reads them across every conversation.

Get started free See pricing

Give your AI a memory

Write once. You both remember.