How to use Claude Code in 2026: complete setup guide

Claude Code productive in under an hour: install, configure MCP, learn subagents and slash commands. Here's the exact setup.

Step 1: Install (2 minutes)

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

Sign in with your Claude Pro ($20/mo) or Max ($100/mo) subscription. Pro covers most users.

Step 2: Configure your project (5 minutes)

In your project root, create CLAUDE.md:

# Project: AI Agent Rank

Tech stack: Next.js 15, React 19, TypeScript, Tailwind, Supabase + Drizzle.

## Conventions
- Single quotes, 2-space indent
- Server components by default
- Functional components with hooks

## Avoid
- Class components, jQuery, lodash

Claude Code reads this on every session. Update when conventions evolve.

Step 3: Install MCP servers (10 minutes)

Edit ~/.claude.json:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {"GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_xxx"}
    },
    "linear": {
      "command": "npx",
      "args": ["-y", "@linear/mcp"],
      "env": {"LINEAR_API_KEY": "lin_xxx"}
    }
  }
}

Restart Claude Code. Verify with /mcp list.

For more MCP server options see Best MCP servers in 2026.

Step 4: Learn subagents (15 minutes)

Subagents are Claude Code's killer feature. Create ~/.claude/agents/code-reviewer.md:

---
description: Reviews code changes for bugs, style, and security
---

You are a senior code reviewer. Focus on:
- Logic bugs
- Security issues (SQL injection, XSS, secret leaks)
- Performance regressions
- Style and convention adherence

Be concise. Flag the top 5 issues only.

Now invoke: "Have the code-reviewer subagent check this PR".

Common subagents to create: code-reviewer, test-writer, performance-auditor, security-scanner, docs-writer.

Step 5: Master slash commands (5 minutes)

Built-in commands worth knowing:

• /clear — reset context
• /compact — summarize and compress context
• /mcp — manage MCP servers
• /agents — list available subagents
• /cost — see current session token spend

Create custom slash commands in ~/.claude/commands/.

Step 6: Three productive workflows (20 minutes)

Workflow 1: Bug-to-PR

1. claude-code in your repo
2. "Read the Linear ticket LIN-1234 and implement the fix"
3. Claude reads, plans, writes, tests, commits
4. You review and push

Workflow 2: Refactor with review

1. "Refactor the auth module to use the new session API"
2. After Claude finishes, "Have code-reviewer subagent check this"
3. Address feedback; commit

Workflow 3: Investigation

1. "Why is the deploy timing out? Look at git log, recent commits, and the build config"
2. Claude investigates with tool calls
3. Returns a diagnosis with evidence

Three Claude Code mistakes to avoid

1. Skipping CLAUDE.md. Without it, the agent doesn't know your codebase and produces generic patterns.

2. Not using subagents. A single agent doing everything is worse than specialized subagents collaborating.

3. Ignoring /clear between tasks. Context bloats; old tasks pollute new ones. Clear between unrelated work.

The verdict

Claude Code in an hour: install → CLAUDE.md → MCP servers → subagents → slash commands. After setup, it's the most powerful terminal-native coding agent in 2026.

For comparisons see Claude Code vs Cursor, Claude Code vs Codex CLI, and Best coding agents in 2026.