Claude Code

Autonomous Terminal Agent

Claude Code runs in your terminal as an autonomous coding agent that reads your codebase, edits files, and executes commands without constant human input. It talks directly to model APIs without requiring a backend server or remote code index, and asks for permission before making changes.

VS Code, Cursor, Windsurf & JetBrains Extensions

Native extensions for VS Code (also compatible with Cursor and Windsurf) and JetBrains IDEs bring Claude Code into your development environment with inline diffs, a dedicated sidebar panel, real-time code change visibility, @-file mentions with line ranges, and conversation history.

Multi-Agent Orchestration (Subagents)

Spawn multiple Claude Code subagents to work on different parts of a task simultaneously — for example, one building the frontend while another builds the backend API. A lead agent coordinates work, assigns subtasks, and merges results.

Checkpoints & Rewind

The checkpoint system automatically saves your code state before each change. You can instantly rewind to any prior state using Esc×2 or /rewind, restoring the code, the conversation, or both — making it safe to attempt ambitious, wide-scale refactors and delegations.

Hooks & Background Tasks

Hooks automatically trigger shell commands at specific points in Claude Code’s workflow — such as running your test suite after code changes or linting before commits. Background tasks keep long-running processes like dev servers active without blocking the agent’s progress on other work.

Sandboxed Execution

Built on OS-level sandboxing, Claude Code enforces filesystem isolation (Claude only accesses approved directories) and network isolation (Claude only connects to approved servers). In internal usage at Anthropic, sandboxing reduces permission prompts by 84% while keeping autonomous execution secure against prompt injection and data leakage.

MCP Server Integration

Connect external tools, databases, and APIs to Claude Code via Model Context Protocol servers. MCP gives the agent access to services like Slack, Google Drive, Jira, GitHub, and custom internal tooling, letting it pull context directly into development workflows without manual copy-paste.

Plugins

Plugins are custom collections of slash commands, subagents, MCP servers, and hooks that install with a single command via /plugin. Teams can share and install plugins to standardise agentic development workflows across an engineering organisation.

Claude Code on the Web & Desktop

Run Claude Code in an isolated cloud sandbox at claude.ai/code without local installation, or in the Claude desktop app to manage multiple parallel tasks, review visual diffs, and monitor PR status from one place. Web sessions execute in a secure environment where sensitive credentials are never inside the sandbox.

CLAUDE.md Project Context

Project instructions live in a CLAUDE.md file in the project root. Claude Code reads this file at the start of every session, updates it as it works through issues, and uses it to enforce coding standards, architecture decisions, and review checklists consistently across all sessions.

GitHub & GitLab Integration

Claude Code integrates natively with GitHub and GitLab to read issues, write code, run tests, commit changes, and submit pull requests. In CI/CD, automate code review and issue triage with GitHub Actions or GitLab CI/CD pipelines.

Scheduled & CLI Automation

Run Claude Code on a schedule for repeating tasks — morning PR reviews, overnight CI failure analysis, weekly dependency audits, and post-merge doc syncs. Cloud scheduled tasks run on Anthropic-managed infrastructure and keep running when your computer is off. The CLI follows the Unix philosophy and composes with pipes and other tools.

Auto Mode

Auto Mode lets Claude Code make permission decisions autonomously using a built-in classifier that reviews each tool call before it executes. Safe actions proceed automatically; potentially destructive actions are blocked or escalated for approval. This provides a safer middle ground for long-running autonomous tasks — fewer interruptions than default mode, less risk than skipping permissions entirely.

Agentic Codebase Search

Claude Code uses agentic search to autonomously map and understand your entire codebase without requiring manual context file selection. It identifies relevant files, traces dependencies, and builds a working model of the project architecture before acting — enabling accurate, coordinated changes across the full codebase.