How to set up Claude Code is a practitioner's guide to getting 📝Claude Code running and configured for real work — not just installed, but set up with the memory, permissions, and workflow patterns that make it productive from day one. This guide reflects how someone running a 📝57-agent system actually configures the tool.

Installation

npm install -g @anthropic-ai/claude-code

Requires Node.js 18+ and an 📝Anthropic API key (or a Claude Pro/Max subscription for direct auth). Run claude in your terminal to start. First launch walks you through authentication.

The First File: CLAUDE.md

Before writing any code, create your project's 📝CLAUDE.md file. This is the single highest-leverage action in any Claude Code setup. See: 📝CLAUDE.md as Infrastructure.

# Project: [Your Project Name]
## Architecture
- [Framework, language, key dependencies]
- [File structure overview]
## Development Rules
- [Branching strategy]
- [Testing requirements]
- [Deployment rules — e.g., "never push to main without asking"]
## Conventions
- [Code style, naming, patterns you follow]

Claude Code reads CLAUDE.md at three levels: ~/.claude/CLAUDE.md (global), project root, and subdirectories. Start with a project-level file. Add a global file when you notice patterns repeating across projects.

Permission Configuration

Claude Code asks permission before running commands. For trusted operations, pre-approve them in ~/.claude/settings.json:

{
  "permissions": {
    "allow": [
      "Read", "Edit", "Write", "Glob", "Grep",
      "Bash(git status)", "Bash(git diff)",
      "Bash(npm test)", "Bash(npm run build)"
    ]
  }
}

Start conservative. Add permissions as you build trust. This mirrors 📝Collaborative Augmentation: the system earns expanded authority through demonstrated competence.

Connect Your Knowledge System

If you use a knowledge platform like 📝MythOS, connect it via 📝MCP:

claude mcp add mythos \
  --scope user \
  https://mythos.one/api/mcp \
  --header "x-mythos-key: YOUR_API_KEY" \
  --header "x-mythos-username: YOUR_USERNAME"

This gives Claude Code access to your knowledge library across every project. See: 📝MythOS MCP.

Key Workflows

Plan Mode

For non-trivial tasks, use plan mode — Claude Code reads the codebase, proposes an approach, and waits for approval before writing code:

Type your task description
Claude enters plan mode and proposes an approach
Review and approve (or redirect)
Claude executes the plan

This is 📝agentic coding at its best: the agent plans, you verify, the agent builds.

Sub-Agents

For complex tasks, Claude Code spawns specialized sub-agents that work in parallel:

Research agents explore the codebase while you continue the main conversation
Background agents handle long-running tasks without blocking
Each sub-agent gets its own context window, keeping the main conversation clean

Hooks

Shell commands that fire on tool calls, notifications, or other events. Use hooks for custom automation — running linters after edits, posting to Slack after commits, or triggering builds after file changes.

The Practitioner Setup

After running Claude Code across multiple production applications, here's the configuration that works:

Global CLAUDE.md — collaboration patterns, workflow principles, correction history (synced from 📝MythOS)
Project CLAUDE.md — architecture decisions, deployment rules, branching strategy
MCP connections — MythOS for knowledge, Playwright for browser automation, project-specific servers as needed
Permission allow-list — pre-approve read/write/grep operations. Require approval for git push, deploy, and destructive operations
Verification habit — always run tests, check logs, review diffs before accepting. See: 📝Building with Claude Code as a Non-Engineer

Every Claude Code setup guide I've read stops at installation. "Run npm install, type claude, you're done." That's like saying you've set up a workshop because you bought a hammer. The real setup is the CLAUDE.md file, the MCP connections, the permission model, and the verification habits. Those are the things that make Claude Code productive versus impressive-but-dangerous.

My setup evolved over a year of daily use. The CLAUDE.md syncs from MythOS so it's always current. The permissions are tuned to pre-approve the safe stuff and gate the risky stuff. The MCP connections give every session access to my full knowledge base. That's not a tool setup. That's 📝Augmentation Stack infrastructure.