Skip to content

Setup

Connect Lodekeep to Claude Code

Claude Code speaks to the hosted connector directly. No npm install, no key to paste: it signs you in with OAuth.

Terminal, about thirty seconds. Requires the Max or Team plan on Lodekeep: coding-agent clients are a paid surface. Chat clients (Claude, ChatGPT, Grok) work on every plan.

  1. One command. The --scope user part matters: without it Claude Code registers the server for the current folder only, and your memory would disappear the next time you open a different project.
  2. terminal
    claude mcp add --scope user --transport http lodekeep https://lodekeep.com/mcp
  3. Then sign in. Either run claude mcp login lodekeep from the shell, or type /mcp inside a session, pick lodekeep and choose Authenticate.
  4. check it worked
    claude mcp list
  5. You want to see lodekeep marked connected. If it says it needs authentication, run the login step above. On a headless or SSH machine add --no-browser to print the URL instead of opening one.
  6. To share it with a repo instead, use --scope project and Claude Code writes a .mcp.json you can commit. Already added Lodekeep on claude.ai? Then it is here already, nothing to run.

Check it worked

Say “Remember this: my test phrase is gold vein.” Then open a different AI you have connected and ask “what is my test phrase?” If it answers gold vein, the same memory is reaching both.

Questions about Claude Code

What does it cost?
Claude Code is a coding-agent client, so it needs Lodekeep Max or Team. Chat clients (Claude, ChatGPT, Grok) work on every plan including free.
How do I install it?
One command: npx lodekeep@latest connect --client claude-code. It merges Lodekeep into your config without touching your other servers.
Does it work across projects?
Yes. Memory is organised into projects and each one is individually included in or excluded from recall, so a work project and a side project stay separate.
Can I use it with Cursor at the same time?
Yes. The same memory is readable from every connected client, so a decision recorded in Claude Code shows up in Cursor.
What happens when I correct something?
Correcting a fact supersedes the old entry, and the stale version stops being recalled. Newer notes also outrank older ones.
What if it does not know the answer?
It returns nothing rather than guessing. There is no keyword fallback, so an unknown question does not produce noise.

Full guide on the setup page. Stuck? Email [email protected].