Customizing Your Setup

This kit is opinionated by default, customizable by design.

---

Adding MCP Servers

Edit ~/.claude/mcp-servers.json to add more servers:

{
  "your-server": {
    "command": "npx",
    "args": ["-y", "your-mcp-server@latest"],
    "description": "What it does"
  }
}

Popular additions:

• @anthropic-ai/mcp-server-memory — if you want dedicated persistent memory
• firecrawl-mcp-server — advanced web scraping with JS rendering
• @anthropic-ai/mcp-server-github — GitHub API integration

Rule of thumb: Every MCP server costs ~500-2000 context tokens just to exist. Only add servers you'll use in most sessions.

---

Modifying Hooks

Hooks live inside ~/.claude/settings.json under the hooks key.

Disable a hook

Remove it from the array, or comment it out (JSON doesn't support comments, so you'll need to remove it).

Add a new hook

{
  "PostToolUse": [
    {
      "matcher": "Edit|Write",
      "hooks": [
        {
          "type": "command",
          "command": "your-command-here"
        }
      ]
    }
  ]
}

Hook exit codes

• exit 0 — allow the action to proceed
• exit 1 — hook failed (action still proceeds, error is logged)
• exit 2 — block the action and show stderr to the user

Useful hook ideas

• Run eslint --fix after JS/TS edits
• Run cargo fmt after Rust edits
• Block edits to package-lock.json (force use of npm install)
• Log all Bash commands to a file for audit

---

Adding Agents

Create a .md file in ~/.claude/agents/:

---
name: my-agent
description: What it does in one line
model: sonnet  # or opus, haiku
tools:
  - Read
  - Grep
  - Glob
disallowedTools:
  - Edit
  - Write
---

Your agent's system prompt goes here.

Key settings:

• model: haiku — fast and cheap for simple tasks
• model: opus — deep reasoning for complex analysis
• model: sonnet — best balance for most agents
• isolation: worktree — runs in a git worktree (isolated copy of repo)

---

Modifying Permissions

In ~/.claude/settings.json:

{
  "permissions": {
    "allow": ["Bash(git *)"],
    "deny": ["Bash(rm -rf *)"]
  }
}

Pattern syntax:

• Bash(git *) — allow any git command
• Bash(npm test*) — allow npm test and npm test:unit etc.
• Read(~/.ssh/*) — allow/deny reading SSH keys

Less restrictive (for personal projects):

"allow": ["Bash", "Read", "Write", "Edit"]

More restrictive (for work/sensitive projects):

"allow": ["Bash(git status)", "Bash(git diff)", "Bash(git log)", "Read", "Glob", "Grep"]

---

Changing the CLAUDE.md Template

The starter CLAUDE.md is deliberately minimal. Fill in the sections for each project:

## Stack
TypeScript, Next.js 15, PostgreSQL, Prisma, Tailwind

## Commands
- `npm run dev` — start dev server on port 3000
- `npm test` — run vitest
- `npm run db:push` — push schema changes

## Architecture
- app/ — Next.js app router
- components/ — React components (PascalCase.tsx)
- lib/ — shared utilities
- server/ — tRPC routers

## Critical Constraints
- Always use server actions for mutations, never API routes
- Use Prisma transactions for multi-table writes
- Never import server code in client components

---

Profile Layering

Profiles stack. essential is always applied first, then your chosen profile overlays on top.

To create your own profile:

1. Create profiles/my-profile/
2. Add any of: settings.json, hooks.json, mcp-servers.json, keybindings.json, agents/*.md
3. Run ./install.sh --profile my-profile

Settings are merged (your profile adds to essential), not replaced.