Setting up Claude Code

A post from the future!

You may have noticed this post is dated in the future. I do this to keep myself on a regular writing schedule and to mask the flurries of activity I'm prone to.

Please feel free to read and share as you normally would.

I've recently been utilising and customising Claude Code more - rather than just sticking with what comes out of the box. With more and more of my colleagues and friends adopting AI (and, specifically, Claude), I thought I would share my go-to config when setting it all up.

Terminal

Claude Code on the terminal is where I spend most of my time - I find it is more powerful than the Code element of the desktop app. Once Claude is installed, these are the settings I reach for:

  • /model: Sonnet - I've found Sonnet to be a good reliable workhorse without burning all the tokens
  • /effort: High - Effort was explained to me like "motivation" - High is a good, solid balance
  • /advisor: Opus - This is a killer feature - it does a lot of the thinking in Sonnet and then checks with Opus
  • /config - Some settings I change from the default
    • Default permission mode: Plan - Means I don't have to remember to switch, sometimes the plan gives you a solution without needing to burn more tokens implementing it
    • Push when actions required: true

MCP / CLI tools

The next thing I lean on is MCPs and CLIs to expand the capabilities of Claude

MCPs

Model Context Protocol (MCP) servers allow LLMs and AI tools to connect to live data to get more context for the task.

  • ClickUp - This is our project management tool and allows Claude to get the full context of a task
  • GitLab - We use GitLab for our projects so allowing Claude to access Work Items (issues) and MR feedback helps speed up the feedback loop

CLIs

With certain CLI tools, Claude is able to perform actions more efficiently which saves tokens

  • GitLab CLI - Access issue and merge request information over a simplified CLI
  • Playwright - Preferred over the Playwright MCP

Skills & Plugins

Not so much a config or command in itself, but if you find yourself repeating particular instructions (e.g. how to make a PHPUnit test or code something in a particular way) consider making a skill.

The best way to do this is to ask Claude itself:

"Make a skill that helps me write consistent..."

Don't forget to specify if it should be a project or user skill.

You can use the skill creator plugin to help you create skills:

  1. Type /plugin
  2. Search for skill-creator, press "Enter" and install it
  3. Type /reload-plugins to register it
  4. Type /skill-creator:skill-creator with a description to get you started

It's worth looking through the /plugin list for plugins which may help with the type of coding you do.

Learning

Master.dev features a free course on Claude. It doesn't go into deep complex architecture, but gives a great overview to help you understand Claude better and how to get the best of it

Conclusion

And that's it, I'll update the post as I find more customisations and tweaks. For now, this is how I set up my Claude code. If you've got this far, you might be interested to read my true feelings about AI.

View this post on Github

You might also enjoy…

Mike Street

Written by Mike Street

Mike is a CTO and Lead Developer from Brighton, UK. He spends his time writing, cycling and coding. You can find Mike on Mastodon.