Claude Code Features
The complete feature surface of 📝Claude Code, organized by 📝Anthropic's own documentation categories and compiled from its published docs. Features with a dedicated memo link to it; the rest are listed by capability. Terms are defined separately in the 📝Claude Code Glossary. For the full collection, browse 🏷️#feature-of-claude-code.
Surfaces
- Terminal CLI — the full-featured command line for editing files, running commands, and managing a whole project
- VS Code extension — inline diffs, @-mentions, plan review, and conversation history inside the editor
- JetBrains plugin — IntelliJ, PyCharm, WebStorm and others, with interactive diffs and selection context sharing
- Desktop app — parallel sessions with git isolation, visual diff review, an integrated terminal, and app previews
- Claude Code on the web — browser sessions with no local setup, for long tasks and repos you do not have locally
- Claude Code on mobile — start, monitor, and steer tasks from the Claude app for iOS and Android
- Development containers — run Claude Code inside a dev container for a consistent, isolated environment across a team
Instructions and Memory
- 📝CLAUDE.md — a markdown file read at the start of every session for standards, architecture, and context
- Auto memory — accumulates build commands and debugging learnings across sessions without you writing anything
- 📝Claude Code Memory — file-based persistence that survives across conversations
- The .claude directory — where CLAUDE.md, settings, hooks, skills, commands, subagents, workflows, and rules live
- Nested CLAUDE.md files — per-package instructions so a monorepo session stays focused on the code in hand
Extending Claude Code
- 📝Skill — packaged repeatable workflows a team can share, plus custom commands and bundled skills
- 📝Slash Command — a typed command invoking a built-in or a skill registered via plugin
- 📝Hook — shell commands that run automatically on file edits, task completion, or input requests
- 📝Claude Code Plugin — skills, agents, hooks, and MCP servers bundled as one installable unit
- Plugin marketplaces — discover, install, and distribute prebuilt plugins across teams and communities
- 📝MCP Server — connects to external tools and data over 📝Model Context Protocol (MCP)
- Output styles — alternative system prompts that adapt Claude Code for work beyond software engineering
- Keyboard shortcuts — customize keybindings through a configuration file
Agents and Orchestration
- 📝Claude Code Subagents — specialized agents with isolated context for task-specific work, running in the background by default
- 📝Claude Code Agent View — dispatch and manage many sessions from one screen, showing which ones need your input
- Agent teams — coordinate multiple instances with shared tasks, inter-agent messaging, and central management (see 📝Claude Code Swarm Mode)
- 📝Claude Code Dynamic Workflows — a rerunnable script orchestrating many subagents for audits, migrations, and cross-checked research
- 📝Claude Code Projects — one ongoing conversation where a coordinator runs related tasks as parallel cloud threads
- 📝Worktree — isolated git worktrees so parallel sessions never collide on the same files
- 📝Claude Code Cross-Session Messaging — list and message your other sessions on one machine, or across machines via Remote Control
Session Control
- 📝Plan Mode — propose an approach before any writes, and proceed only on your approval
- 📝Auto Mode — a classifier handles routine decisions against repos, buckets, and domains your org trusts
- 📝Permissions — fine-grained allow, ask, and deny rules over tools, commands, and MCP calls
- Permission modes — cycle how often Claude asks before editing files or running commands
- 📝settings.json — global and project-level configuration, plus environment variables
- Sessions — name, resume, branch, and switch conversations, including resuming the session that opened a pull request
- Checkpointing — track, rewind, and summarize edits and conversation to manage session state
- 📝/goal — sets a completion condition and keeps working across turns until it holds
- Interactive mode — keyboard shortcuts, input modes, and the interactive features of a live session
- Voice dictation — speak your prompts with hold-to-record or tap-to-record
- Status line — a custom bar monitoring context usage, cost, and git status
- Screen reader mode — VoiceOver and NVDA support, plus magnifier, reduced-motion, and colorblind-friendly themes
- Fullscreen rendering — flicker-free output with mouse support and stable memory in long conversations
Context and Cost
- 📝context window — what loads automatically, what each file read costs, and when rules and hooks fire
- 📝Compaction — automatic summarization as a conversation approaches the context limit
- Prompt caching — managed automatically, with a model switch triggering one slow uncached turn
- 📝ENABLE_PROMPT_CACHING_1H — the extended one-hour cache for important but less frequently hit context
- 📝Model — model configuration including aliases such as opusplan
- Fast mode — faster Opus responses, toggled per session
- Cost management — token tracking, team spend limits, and the levers that actually reduce spend
Automation and Scheduling
- 📝Claude Code Routines — cloud runs on a schedule, on API calls, or on GitHub events, working while your machine is off
- Desktop scheduled tasks — recurring runs on your own machine with direct access to local files and tools
- /loop and cron tools — repeat a prompt within a session, poll for status, or set a one-time reminder
- 📝Headless Mode — run Claude Code programmatically from the CLI, Python, or TypeScript
- Deep links — open a session from a URL, so a click in a runbook, alert, or dashboard lands in the right repo
Remote Work and Handoff
- Remote Control — continue a local session from your phone, tablet, or any browser
- Cloud sessions — move work between web and terminal with the cloud and teleport flags
- Dispatch — message a task from your phone and open the desktop session it creates
- Cloud environments — configure network access levels, environment variables, setup scripts, and caching
- Self-hosted environments — run cloud sessions on infrastructure you control, with your own runners and identity
Integrations
- Chrome — test web apps, debug with console logs, automate form filling, and extract data from pages
- Computer use — open apps, click, type, and see your screen on macOS from the CLI
- GitHub Actions — respond to mentions, automate tasks, and turn issues into pull requests
- GitLab CI/CD — the same automation on GitLab pipelines
- Code Review — automated PR review catching logic errors, security vulnerabilities, and regressions
- ultrareview — a deep multi-agent cloud review that finds and verifies bugs before you merge
- Claude Tag — brings Claude into your team's Slack channels
- Slack — delegate coding tasks from a workspace; being retired for Team and Enterprise in favor of Claude Tag
- Artifacts — publish session output as a live, interactive page on claude.ai, private or shared
- GitHub Enterprise Server — connect web sessions, code review, and plugin marketplaces to a self-hosted instance
Security and Isolation
- Sandboxed Bash tool — filesystem and network isolation for safer, more autonomous execution
- Sandbox environments — pick an isolation level across the sandboxed tool, sandbox runtime, dev containers, Docker, and VMs
- Claude Security plugin — scans a codebase for vulnerabilities and turns findings into patches you review and apply
- security-guidance plugin — has Claude review its own changes for vulnerabilities and fix them in the same session
- Zero data retention — available to qualified Claude for Enterprise accounts, with a documented disabled-feature set
Agent SDK
- Agent SDK — build production agents with Claude Code as a library, in TypeScript or Python
- Custom tools — define your own functions as tools through an in-process MCP server
- Structured outputs — return validated JSON from agent workflows using JSON Schema, Zod, or Pydantic
- SDK sessions — persist, resume, and fork conversation history, optionally mirrored to S3, Redis, or your own backend
- File checkpointing — track file changes during a run and restore files to any previous state
- Tool search — scale an agent to thousands of tools by loading only what the task needs
- Observability — export traces, metrics, and events to your backend over OpenTelemetry
- SDK hosting — subprocess architecture, session persistence, scaling, and multi-tenant isolation
Enterprise Deployment
- Managed settings — centrally enforce policy, model access, and configuration across an organization
- Server-managed settings — deliver that configuration from a server without device management infrastructure
- Claude apps gateway — a self-hosted gateway with SSO, per-group model access, spend limits, and OTLP telemetry
- Other LLM gateways — route Claude Code through a gateway your organization already runs
- Amazon Bedrock — run Claude Code against Bedrock with its own setup and IAM configuration
- Claude Platform on AWS — the Anthropic-operated API with AWS authentication and Marketplace billing
- Google Cloud's Agent Platform — formerly Vertex AI, with its own setup and IAM configuration
- Microsoft Foundry — run Claude Code through Foundry
- Managed MCP — restrict which MCP servers users can add or connect to, with allowlists and denylists
- Team analytics — usage metrics, adoption tracking, and engineering velocity in a dashboard
- OpenTelemetry monitoring — export Claude Code telemetry to your observability backend
- Enterprise network configuration — proxy servers, custom certificate authorities, and mutual TLS
- Feature availability matrix — which features work on each subscription plan, console, and cloud provider
