Claude Code Architecture and Features Guide

At the core of Claude Code is a single, unified execution engine that powers all client surfaces. Whether you run commands in your terminal CLI, trigger inline diffs inside VS Code or JetBrains, run side-by-side sessions in the Desktop app, or launch background tasks on the web (claude.ai/code), your project configuration remains strictly identical. Learn more about advanced programmatic execution in our Claude Agent SDK Guide.

Because every surface connects to the same underlying engine, project instructions in CLAUDE.md, saved auto-memory insights, registered skills, and connected MCP servers are automatically shared. Subscriptions (Claude Pro/Team/Enterprise) or Anthropic Console API keys authenticate all surfaces, while the Terminal CLI and VS Code extension also support third-party LLM providers.

Claude Code supports multiple installation pathways tailored to developer environments across macOS, Linux (Debian, Fedora, RHEL, Alpine), WSL, and native Windows. Choosing the correct installer determines how background updates and shell dependencies are handled. For isolated sandbox execution, see our Sandboxing & Isolation Guide.

Native Installers vs. Package Managers

The native installer is the recommended installation method because it includes automatic background updates:

• macOS / Linux / WSL: curl -fsSL https://claude.ai/install.sh | bash
• Windows PowerShell: irm https://claude.ai/install.ps1 | iex
• Windows CMD: curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

1# Homebrew Install (macOS / Linux)2brew install --cask claude-code           # Follows stable channel (~1 week delay)3brew install --cask claude-code@latest    # Follows latest channel (instant updates)4 5# Upgrade Homebrew & WinGet installs manually6brew upgrade claude-code7winget upgrade Anthropic.ClaudeCode

On native Windows, installing Git for Windows is strongly recommended so Claude Code can utilize its native Bash tool instead of defaulting to PowerShell. On first launch inside any project directory (cd my-project && claude), you are prompted to log in. Setting the ANTHROPIC_API_KEY environment variable bypasses interactive login prompts.

Claude Code automates tedious development workflows such as writing unit tests for legacy code, fixing linting errors project-wide, resolving Git merge conflicts, updating dependencies, and drafting release notes. When given a bug report or stack trace, it traces the problem through the codebase to pinpoint root causes and implement verified fixes.

Customization via CLAUDE.md, Skills, and Hooks

Projects are customized using four extensible mechanisms:

• CLAUDE.md: A project-root markdown file read at session startup containing coding standards, architectural rules, and build commands.
• Auto-Memory: Automatically saved insights (such as build quirks or test flags) that persist across sessions without manual documentation.
• Skills: Packaged, reusable team workflows triggered by slash commands (e.g., /review-pr or /deploy-staging).
• Hooks: Shell commands triggered before or after agent actions—such as auto-formatting files after edits or running linters prior to commits.

Model Context Protocol (MCP)

Through MCP, Claude Code connects to external enterprise services—allowing it to read design documents from Google Drive, update Jira issues, fetch context from Slack, or query custom internal databases.

For complex, multi-faceted tasks, Claude Code supports sub-agent delegation. A lead agent breaks down high-level objectives, spawns specialized sub-agents to tackle isolated subtasks in parallel, and merges the results into a cohesive solution.

Scheduled Execution: Routines vs. Desktop Loops

Automation tasks can be scheduled to run locally or in the cloud:

1# Pipe log output directly into Claude Code (Unix Philosophy)2tail -n 200 production.log | claude -p "Notify Slack if critical anomalies exist"3 4# Perform bulk security review across modified files5git diff main --name-only | claude -p "Review changed files for security vulnerabilities"6 7# Repeat prompt polling inside terminal session8/loop 5m "Check deployment status and report errors"

• Cloud Routines: Created on the web, in the Desktop app, or via /schedule. They run on Anthropic infrastructure even when your computer is powered off.
• Desktop Scheduled Tasks: Run on your local hardware with direct access to local filesystem paths and private network tools.

Cross-Device Teleportation & Integrations

With Remote Control and Dispatch, developers can monitor local sessions from mobile devices (iOS/Android). Running claude --teleport pulls web or mobile tasks back into your local terminal. The /desktop command seamlessly transitions a terminal session into the Desktop app for visual diff review, while Slack integrations allow users to @Claude a bug description and receive a working PR in return.

To consolidate your understanding of Claude Code, configure a local project environment with a custom CLAUDE.md and a documented CLI verification command. Keep the hook event names and settings schema aligned with the current Anthropic documentation; the exact event payload should be tested before enabling automatic writes.

1mkdir -p my-claude-project && cd my-claude-project2cat << 'EOF' > CLAUDE.md3# Project Guidelines4- Architecture: Modular TypeScript Next.js App Router5- Testing: Run the repository verification command before committing6- Formatting: Format modified files before review7EOF8 9# Non-interactive audit; inspect the output and exit status10git diff --name-only | claude -p "Review these changed files against CLAUDE.md and report failures"

Intentional failure: add a formatting violation or failing test. Recovery: fix the violation, rerun the project check, and pass only when the command exits successfully and the diff contains no unreviewed changes.