This is a list of the various things I’ve done to my claude code setup. A lot of it is not well documented, so I’m pulling it together as a master list here.

# Plan mode

Always hit shift-tab twice when starting a new session to have Claude plan first, then execute. Do the same when steering significantly, or starting a new task.

# Background tasks

Add export ENABLE_BACKGROUND_TASKS=1 in your shell profile to allow claude to run long-running tasks in the background.

# MCP servers

MCP servers can be local or remote. Installing them to claude code is either done by adding a block like:

"mcpServers": {
  "puppeteer": {
    "type": "stdio",
    "command": "npx",
    "args": [
      "-y",
      "u/modelcontextprotocol/server-puppeteer"
    ],
    "env": {}
  }
}

to Claude Desktop config and running claude mcp add-from-claude-desktop, or running a command like claude mcp add --transport http Ref https://api.ref.tools/mcp?apiKey=YOUR_API_KEY.

Skip the Claude Desktop step (unless you also want Claude Desktop to have those MCPs) and add the code block directly to ~/.claude.json. When installing via the CLI, add -s user to populate the same part of the config file and use those MCPs from anywhere.

# Documentation

There are a lot of recommendations for the context7 MCP server, but I find Ref faster, more reliable, and token-efficient. There’s a generous free tier¹, so it’s worth a try. Here’s a side-by-side comparison on finding docs for Apple’s Foundation Models framework:

# Teleoperation

I still use my custom mobile dev setup, but VibeTunnel is far easier to set up, and has a bunch of nice UX improvements like mouse mode and session management. A setup like this let’s me easily switch from my laptop to my phone, and continue working on stuff AFK. Having a server helps, but you can run it off a laptop too²

# Optimisation

There’s a lot of buzz about frameworks like SuperClaude but I’m pretty sure it doesn’t work, and significantly pollutes the context window. Rather than installing and learning a framework like this, I prefer to:

• optimise my global and project-level CLAUDE.md
• add custom commands for things I do often
• add MCPs sparingly

Configuring and optimising these tools is still very vibe-based, so it’s surprisingly easy to fool yourself into thinking something works and adding layers of complexity you don’t understand and don’t need. Keep it simple, stupid!