What We'll Cover Today

Claude Code: Your AI Pair Programmer

• Official Anthropic CLI for Claude 4.5 Sonnet.
• Native integration with your local development environment.
• Works where you work: Terminal + VSCode support.
• Designed from the ground up specifically for engineers to be fast and context-aware.

Key Capabilities

Direct Codebase Access

Reads and understands your project files without copy-pasting.

Complete Project Context

Gains a holistic understanding of your entire repository.

Command Execution

Can run bash commands to check status, run tests, or install dependencies.

File Editing

Applies suggested changes directly to your files with your approval.

MCP Tool Integration

Connects to external tools like Kubernetes, GitHub, and databases.

Specialized Agent System

Delegates complex tasks to a team of specialized subagents.

What You'll Need

Your GCP Sandbox Environment

Installation Steps LIVE DEMO

curl -fsSL https://claude.ai/install.sh | bash

export CLAUDE_CODE_USE_VERTEX=1
export CLOUD_ML_REGION=global
export ANTHROPIC_VERTEX_PROJECT_ID=mo-sandbox-yourname

gcloud auth application-default login

claude

Claude Code CLI Basics

Terminal Power in Action LIVE DEMO

claude "Summarize this project's purpose and tech stack. What's the entrypoint file?"

claude "Analyze error.log. What is the root cause of the frequent NullPointerException?"

claude "Update the Terraform config to increase the instance count from 3 to 5 and apply the changes"

claude "Refactor 'calculate_totals' in utils.py for better performance and add comments"

Powerful Slash Commands

/agents

/mcp

/permissions

Claude Code Memory

# "What coding style do we use?"

/memory

Claude Code in Your Favorite IDE

Installation

1. Open the Extensions view (Ctrl+Shift+X).
2. Search for "Claude Code".
3. Click "Install" on the official extension by Anthropic.
4. Reload VSCode and log in via the command palette.

Integrated Development Experience

Recommended VSCode Workflow

Extending Claude's Capabilities

MCP Server Ecosystem

Kubernetes MCP

Interact with your k8s cluster: check pods, read logs, apply manifests.

Atlassian MCP

Interact with Jira and Confluence to manage issues and documentation.

GitHub MCP

Manage repos, issues, and pull requests directly from the CLI.

Slack MCP

Send messages, read channels, and interact with your team.

Custom MCP

The true power: Build your own MCP server to connect to any internal API or service!

Setting Up MCP

claude mcp add kubernetes

claude mcp list

Managing Secrets in MCP Servers

{
  "apiKey": "sk-1234567890abcdef"  ❌
}

{
  "env": {
    "API_KEY": "${GITHUB_TOKEN}"
  }
}

Example: Kubernetes Integration LIVE DEMO

claude "Using the kubernetes tool, list all pods in the 'production' namespace 
that are in a CrashLoopBackOff state, then get the logs for the first one you find."

> Executing: kubectl get pods -n production | grep CrashLoopBackOff
> Executing: kubectl logs -n production pod-xyz-12345

Specialized Agents for Complex Tasks

• A team of task-specific agents working under the main Claude agent.
• Enables autonomous multi-step execution for complex problems.
• You can delegate a high-level goal, and the agent system will figure out the steps to achieve it.

Available Subagents

General-purpose Agent

Breaks down complex research and multi-step tasks into smaller, executable steps.

Code-reviewer Agent

Performs a deep static analysis of code, checking for bugs, style violations, and security issues.

Test-runner Agent

Finds, executes, and analyzes the results of your project's test suite.

Custom Agents

Define your own agents with specific goals and capabilities for your team's unique workflows.

Subagent Use Cases

• Complex multi-step tasks: "Figure out why our staging deployment is failing, fix it, and redeploy."
• Deep codebase research: "Find all instances where we use the deprecated 'lib-v1' and create a migration plan to 'lib-v2'."
• Automated pipeline execution: "Run the test suite. If it passes, run the code-reviewer agent on the new changes. If that passes, create a pull request."

Building Your Own Agents

.claude/agents/your-agent-name.md

---
name: code-reviewer
description: Review code for bugs and issues
tools: [Read, Grep, Edit, Bash]
---

Act as a senior engineer reviewing code.
Check for bugs, performance issues, and
adherence to best practices...

Real-World Scenarios

More Real-World Scenarios

Documentation Generation

"Generate a README.md for this microservice. Include a project summary, setup instructions, and API endpoint documentation based on the code."

Infrastructure as Code

"Write a Terraform module to provision a secure S3 bucket with versioning, encryption, and a lifecycle policy to move objects to Glacier after 90 days."

Code Migration

"I need to migrate this project from Python 2 to Python 3. Analyze the codebase and provide a list of necessary changes and the refactored code."

How to Get Better Results

• Be specific and contextual: Instead of "fix this code," say "This code throws a TypeError when the input is null. Please add a check to handle this case."
• Provide examples: Give input and expected output. "For a pod named 'api-server' in namespace 'production', the kubectl command should return its CPU and memory usage."
• Define constraints: Mention tech stack, style guides, or performance needs. "Refactor this using ES6 syntax, and it must not use any external libraries."
• Use extended thinking for complex problems: Start your prompt with "Think:" to enable Claude's extended thinking mode, allowing deeper reasoning for algorithmic challenges, architectural decisions, or complex debugging.
• Iterate and refine: If the first answer isn't perfect, provide feedback and ask for a revision. "That's a good start, but can you also add error logging?"

Critical: Human Judgment Required

• NEVER apply code without a thorough review. You are the final authority.
• Understand every line of code you commit. If you don't understand it, ask Claude to explain it.
• Test locally first. Always validate AI-generated code with your own tests.
• Do not share secrets, API keys, or proprietary data in your prompts.
• Validate for security vulnerabilities. AI can sometimes produce insecure code.

Using AI Wisely

Learn More

Documentation

docs.claude.com/en/docs/claude-code/overview

GitHub

github.com/anthropics/claude-code

Slack Channel

#cloud-engineering

Homework for Next Session

Coming Up Next...