How to Set Up Claude Code: A Practical Guide

TL;DR

• Install via pip install claude-code or your IDE’s marketplace (VS Code, JetBrains, Neovim).
• Configure with .claude/ directory and CLAUDE.md for project-specific rules.
• Use claude suggest, claude fix, and claude test for core workflows.
• Integrate with MCP server for multi-agent collaboration (2026+ feature).
• Manage costs by setting usage limits and enabling "Fast Mode" for large projects.

1. Installation and First Run

CLI Installation

# Python 3.9+ required
pip install claude-code --upgrade
claude --version  # Verify (expected: 1.3.x)

Expected Output:

claude-code v1.3.0 (Anthropic, 2026)

IDE Setup

• VS Code: Install from Marketplace and reload.
• JetBrains: Search "Claude Code" in Plugins (Settings > Plugins).
• Neovim: Add to lazy.nvim:

  {
    'anthropics/claude-code.nvim',
    dependencies = { 'nvim-lua/plenary.nvim' },
    config = function() require('claude-code').setup() end
  }
  

  Copy

First Run

claude auth login  # Opens browser for Anthropic API key
claude suggest "Write a Python function to parse JSON with error handling"

Expected Output:

import json

def safe_json_parse(json_str: str) -> dict | None:
    """Parse JSON string with error handling.

    Args:
        json_str: Input string to parse.

    Returns:
        Parsed dict or None if parsing fails.
    """
    try:
        return json.loads(json_str)
    except json.JSONDecodeError as e:
        print(f"Invalid JSON: {e}")
        return None

2. Configuration

.claude/ Directory

Create a .claude/ folder in your project root with these files:

config.yaml

# .claude/config.yaml
model: "claude-3.5-sonnet"  # Default (options: claude-3-haiku, claude-3-opus)
max_tokens: 4096
rulesets:
  - python_style  # Built-in (see docs for custom rules)
  - security_scanning

CLAUDE.md

# Project-Specific Rules
- Always use `async/await` for I/O operations.
- Prefer `pathlib.Path` over `os.path`.
- Document all public functions with Google-style docstrings.

# Ignore Files
- `**/migrations/*`
- `**/generated/*`

Verify Config

claude config validate

Expected Output:

✓ Valid config (rulesets: python_style, security_scanning)
✓ CLAUDE.md loaded (2 rules, 2 ignores)

3. Key Commands and Workflows

Core Commands

Multi-File Workflow

# Analyze cross-file dependencies
claude analyze --files src/main.py,src/utils.py

# Generate a feature across multiple files
claude suggest "Add OAuth2 login" --output=src/auth/

Expected Output:

✓ Generated src/auth/routes.py (120 lines)
✓ Updated src/config.py (added OAUTH2_CLIENT_ID)
✓ Created tests/test_auth.py (8 tests)

4. MCP Server Integration (2026+)

Claude Code’s Multi-Agent Collaboration Protocol (MCP) lets agents coordinate on tasks.

Start MCP Server

claude mcp start --port=8001

Expected Output:

MCP Server running on ws://localhost:8001
Agents: 1 (default)

Connect Agents

# In another terminal
claude mcp connect --name=frontend --role=react
claude mcp connect --name=backend --role=flask

Assign Tasks

# Broadcast a task to all agents
claude mcp task "Implement user profile page" --files=src/frontend/,src/backend/

Gotcha: MCP requires Claude Team/Enterprise for >2 agents MCP Docs.

5. Hooks and Automation

Git Hooks

Add to .git/hooks/pre-commit:

#!/bin/sh
claude fix --staged  # Auto-fix staged files
claude test --changed  # Run tests on changed files

CI/CD Integration

GitHub Actions

# .github/workflows/claude.yml
name: Claude Code Review
on: [pull_request]

jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: anthropics/claude-code-action@v1
        with:
          command: "claude review --pr=${{ github.event.pull_request.number }}"
          api_key: ${{ secrets.CLAUDE_API_KEY }}

6. Best Practices for Large Codebases

Context Management

• Split Large Files: Use --chunk-size=500 for files >1k lines.
• Summarize Context: Add a SUMMARY.md in large directories:

  # src/services/
  - `auth.py`: Handles OAuth2 and JWT.
  - `payments.py`: Stripe/PayPal integration.
  

  Copy

Performance Tips

• Fast Mode: Reduce latency (but lower accuracy):

  claude suggest --fast "Optimize this query"
  

  Copy
• Cache Context: Reuse context for repeated tasks:

  claude cache save --name=auth_module
  claude suggest --cache=auth_module "Add logout endpoint"
  

  Copy

Security

• Scan Secrets: Enable in .claude/config.yaml:

  security_scanning:
    enabled: true
    severity: high  # Options: low, medium, high
  

  Copy
• Audit Logs: Export logs for compliance:

  claude logs export --format=json > claude_audit.json
  

  Copy

7. Cost Management Tips

Monitor Usage

claude usage

Expected Output:

Usage (May 2026):
- Suggestions: 4,287/5,000 (Free tier)
- Security Scans: 12 (Pro feature)
- MCP Agents: 2 (Team feature)

Reduce Costs

• Set Limits: Add to .claude/config.yaml:

  limits:
    daily_suggestions: 1000
    max_tokens_per_request: 2048
  

  Copy
• Use Haiku: Cheaper model for simple tasks:

  claude suggest --model=claude-3-haiku "Format this JSON"
  

  Copy
• Batch Requests: Group similar tasks:

  claude batch "Add docstrings to all utils/*.py"
  

  Copy

Alternatives Comparison

What's Next?

1. Integrate with MCP: Start a local MCP server (claude mcp start) and connect 2+ agents for collaborative coding.
2. Automate Reviews: Add claude review to your CI/CD pipeline (GitHub Actions/GitLab CI).
3. Custom Rules: Define team-specific rules in .claude/rulesets/ (see Rulesets Docs).

For teams scaling [AI-assisted development](https://hyperion-<a href="/services/coaching-vs-consulting">consulting</a>.io/services/ai-development-training), Hyperion Consulting offers tools and frameworks to integrate Claude Code into enterprise workflows.