[BUG] Claude Code always uses system version of node, rather than project-specific version installed via nodenv

The fix here is to write a .zshrc that includes the aliases and env vars that you want claude to have. Claude can't operate fish shell.

I added the setup to .zshrc and it worked, cheers

@asgeo1 @wolffiex Can you give an example of the fix you described?

@bhrutledge it depends on your setup. the idea is that claude code will start bash or zsh (you can have it check which) and will source the .bashrc or .zshrc in your home directory. any config that you want to claude to have can go there

@wolffiex Thanks, but I'm still not quite following. Happy to read the docs if you can point me to them.

the idea is that claude code will start bash or zsh (you can have it check which)

Do you mean that Claude Code automatically starts a shell (e.g. npm)? I use ZSH. How do I have Claude Code check which shell it's using, and/or switch to the one I prefer?

and will source the .bashrc or .zshrc in your home directory. any config that you want to claude to have can go there

Does the sourcing happen automatically? What "config" would I put in there to ensure that npm uses the current nodenv version? I've already have eval "$(nodenv init -)" in my .zshrc.

I recognize that this issue has been closed, but I just wanted to chime in real quick. In my personal .zsh setup, I'm pretty meticulous and obsessive about its organization and launch speed. So I have it set up such that there is a chpwd hook with which I use to detect the nature of the directory I'm in (or changing to). If it detects ruby, node, python, it loads the appropriate *env (load_rbenv, and load_pyenv, etc) manager. In other words, each shell instance doesn't initialize the world, which leaves poor Claude at a disadvantage.

Is there a way to detect that it's a Claude Agent calling for a new shell, and thus I could choose to load the world for it, out of laziness?

This is what I've come up with for now. Still sucks to have to reinit a new shell for each command.

#!/bin/zsh
# claudenv - Claude Code environment initializer

# Set Claude shell indicator
export CLAUDE_SHELL=true

# Source the full zsh configuration
source ~/.zsh/.zshrc

# Somewhere, in a .zshrc far, far away.
function load_dev_tools() {
  load_rbenv
  load_nvm
  load_pyenv
}

if [[ $CLAUDE_SHELL == true ]]; then
  load_dev_tools
fi

And instruct claude to call source ~/bin/claude && [whatever command]

yes, condition on CLAUDE_CODE existing or equaling 1

Sadly, that does not work for me.

!image

Just wasted an hour on this for anyone who comes across this.

Looks like the root issue in my case is https://github.com/modelcontextprotocol/servers/issues/64 but, the workaround is pretty easy. I updated the PATH env variable and the command to specify the node version directly. Just replace the $(USER) with your username and the node version/path if applicable.

"apple-health": {
      "command": "/Users/$(USER)/.nvm/versions/node/v24.3.0/bin/npx",
      "args": ["@neiltron/apple-health-mcp"],
      "env": {
        "HEALTH_DATA_DIR": "/Users/$(USER)/Documents/AppleHealth",
        "PATH": "/Users/$(USER)/.nvm/versions/node/v24.3.0/bin:/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:/usr/sbin:/sbin"
      }
    },

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.