Claude Code

Learning Objectives

• Understand how Claude Code's terminal-based agentic approach differs from chat-based coding assistants
• Identify the key capabilities that make Claude Code effective for complex, multi-step development tasks
• Evaluate when to use Claude Code versus alternatives like OpenAI Codex, Cursor, or Gemini CLI

What Is Claude Code?

Claude Code is Anthropic's command-line coding agent. Unlike browser-based coding assistants, Claude Code runs directly in your terminal — it has full access to your file system, can read and write files, execute shell commands, run your test suite, manage git branches, and interact with external services through the Model Context Protocol (MCP). It is Anthropic's most powerful tool for software engineering.

Claude Code is powered by Claude Opus 4.7 (for complex reasoning tasks) and Claude Sonnet 4.6 (for faster, routine operations), with automatic model routing based on task complexity. It supports a 1 million token context window (GA March 2026), allowing it to reason over very large codebases, and uses a persistent memory system via CLAUDE.md project files that store coding conventions, architecture decisions, and project-specific instructions across sessions.

Claude Code supports sub-agents — ephemeral workers that handle parallel subtasks with a summary returned to the parent — and Agent Teams, where multiple coordinated Claude Code instances share a task list and message each other directly. The Claude Agent SDK (Python and TypeScript) lets developers build custom agents with built-in file operations, shell commands, web search, and MCP integration. Apple's Xcode 26.3 integrated the Claude Agent SDK natively, bringing Claude Code capabilities directly into the IDE.

The practical difference from a chat interface is significant: Claude Code does not generate code snippets for you to copy — it directly edits your files, creates new ones, runs your build and test commands, stages changes in git, and can even create pull requests. You describe what you want, and Claude Code implements it across your actual project.

Pricing and Access

The API (BYOK) option is popular with heavy users — you pay per token with no monthly cap, which can be more cost-effective for developers who use Claude Code extensively throughout the day.

May 2026 Rate Limit Increase

On May 6, 2026, Anthropic doubled Claude Code's five-hour rate limits across Pro, Max, Team, and Enterprise plans, and removed the peak-hours throttle for Pro and Max users. Both changes took effect immediately for existing subscribers, with no price increase. The capacity boost is funded by a parallel announcement: a new compute deal with SpaceX giving Anthropic access to over 300 megawatts and more than 220,000 NVIDIA GPUs at SpaceX's Colossus 1 data center, deployable within one month — the first time SpaceX has been named as a frontier-lab compute partner. Practically, heavy Claude Code users hit hard caps less often and no longer see slowdowns during US business hours.

Token Economics — Microsoft Internal Pilot

A useful real-world data point on Claude Code's token economics surfaced when Microsoft disclosed it is ending its internal employee pilot of Claude Code on June 30 and redirecting staff to GitHub Copilot CLI. The original program was meant to expose project managers and designers to AI coding for the first time, but token consumption ran far ahead of plan — one Microsoft employee reported burning through a monthly token allocation in just over a week, and the program reportedly consumed Microsoft's full-year 2026 internal AI budget within months. The cancellation is procurement economics, not a security or contractual dispute, and notably: when Microsoft's own developers were offered the choice, they picked Claude Code over Copilot, which is what prompted Microsoft to restrict the pilot to protect its own product strategy. For teams evaluating Claude Code seriously, the practical takeaway is to budget for token spend an order of magnitude higher than a Copilot-equivalent baseline and consider API (BYOK) pricing for predictability — heavy daily users routinely outpace flat-rate plan caps.

Core Capabilities

Autonomous Multi-Step Coding

Claude Code handles complex, multi-step development tasks end-to-end. Describe a feature, a bug fix, or a refactoring goal, and it reads your codebase, identifies the relevant files, plans an approach, implements changes across multiple files, and verifies its work by running your tests. It maintains coherence across edits — updating imports, modifying configuration files, and ensuring new code follows your project's existing patterns.

Model Context Protocol (MCP)

Claude Code pioneered the use of MCP servers — standardized connectors that let the agent interact with external tools and services. Out of the box, you can connect Claude Code to Supabase (database queries), Vercel (deployments), Stripe (payment setup), GitHub (PRs and issues), Slack, and thousands of other services. This transforms Claude Code from a coding tool into a full development workflow agent that can deploy code, manage infrastructure, and coordinate with external systems.

Project Memory and CLAUDE.md

Every project can include a CLAUDE.md file at the root — a structured document that tells Claude Code about your project's conventions, architecture, preferred patterns, testing approach, and any rules to follow. Claude Code reads this file at the start of every session, giving it persistent context that survives across conversations. Teams use this to encode institutional knowledge so every team member's Claude Code session follows the same standards.

Voice Mode and Computer Use

Voice Mode (/voice): Push-to-talk voice interaction in 20 languages — describe what you want verbally while your hands stay on the keyboard.

Computer Use: On Mac, Claude Code can operate your desktop — move the mouse, click buttons, interact with browsers and applications. Useful for testing UI changes, interacting with tools that lack APIs, and end-to-end workflow verification.

/loop (Scheduled Tasks): Run prompts on a recurring interval — cron-like scheduling for PR reviews, monitoring, or any periodic task. Example: /loop 5m /review-prs checks for new PRs every 5 minutes.

Git Integration

Claude Code manages git workflows natively. It creates branches, stages specific files, writes commit messages following your project's conventions, and can create pull requests via GitHub MCP. It understands git history and can use git blame and git log to understand who changed what and why — useful context for bug fixes and refactoring.

Strengths

• Terminal-native: Runs in your actual development environment with full file system access — no copying code between browser and editor
• 1 million token context window: Reasons over very large codebases, long files, and complex multi-file relationships
• MCP ecosystem: Connects to databases, deployment platforms, APIs, and external services through standardized protocols
• Project memory: CLAUDE.md files encode project conventions and persist across sessions, ensuring consistent code quality
• Multi-model routing: Automatically uses Opus for complex reasoning and Sonnet for routine tasks, optimizing cost and speed
• IDE integrations: Works within VS Code, JetBrains, and other editors as an extension, not just standalone CLI

Limitations & Considerations

• Terminal comfort required: Best suited for developers comfortable working in the command line — not a visual IDE replacement
• Cloud inference: Code is sent to Anthropic's API for processing — not suitable for air-gapped environments (though code stays local on your machine)
• Usage limits on lower tiers: Free and Pro plans have conversation and token limits that heavy users may hit during long coding sessions
• Review still essential: Like all coding agents, Claude Code can make architectural decisions that need human review — autonomous does not mean unsupervised

Best Use Cases

When to choose alternatives:

• Browser-based development → OpenAI Codex (sandboxed cloud environment, no local setup needed)
• Visual IDE experience → Cursor (AI-native editor with inline suggestions and multi-file editing)
• Google Cloud ecosystem → Gemini CLI (native GCP, Firebase, and Vertex AI integration)
• Privacy-first with no cloud → Local models via Ollama (no code leaves your machine)

Getting Started

1. Install Claude Code: npm install -g @anthropic-ai/claude-code
2. Navigate to your project directory in the terminal and run claude to start a session
3. Create a CLAUDE.md file at your project root with your coding conventions and architecture notes
4. Start with a small task: ask Claude Code to explain part of your codebase or fix a simple bug
5. Try a multi-step task: describe a feature and let Claude Code implement it across multiple files
6. Connect an MCP server (e.g., GitHub) to enable PR creation and issue management from within Claude Code

Key Takeaways

• Claude Code is a terminal-based coding agent that directly reads and writes your project files — not a chatbot that generates snippets to copy
• The MCP ecosystem connects Claude Code to databases, deployment platforms, and external services, making it a full development workflow tool
• Project memory via CLAUDE.md files ensures consistent code quality across sessions and team members
• Best suited for complex multi-file development, debugging, and infrastructure setup; pair with code review for architectural decisions