Skip to content

Code Tab & Claude Code

Last verified: 14 April 2026 | Applies to: Pro, Max, Team, Enterprise

Claude Code is Claude for building and processing. The Code tab in Claude Desktop gives operators coding power through a visual interface. Claude Code CLI is the full terminal version for those comfortable in the command line. Claude Code on the web (claude.ai/code) connects to GitHub repos for cloud-based development. Neither is required to get value from Claude, but both extend what’s possible.

Code tab in Desktop: Click the Code tab in Claude Desktop to start. It’s Claude Code with a visual interface, good for operators who want to build automations, process data, or work with files without opening a terminal.

Claude Code CLI: Install via npm (npm install -g @anthropic-ai/claude-code) or Homebrew. It lives in your terminal, understands your codebase, and helps you code through natural language commands.

Claude Code on the web: At claude.ai/code, Claude connects to your GitHub repos and can clone, edit, test, and propose fixes from a secure cloud sandbox. Useful for reviewing pull requests, fixing bugs, and running tasks asynchronously.

What operators use it for:

  • Building custom automations and internal tools
  • Processing and transforming data at scale
  • Creating plugins programmatically
  • Working with codebases and technical projects
  • Generating scripts for repetitive tasks
  • Resuming previous sessions without losing context
  • Automated code review, fix, and merge workflows
  • Working on parallel branches in isolated worktrees

Session continuity: Claude Code preserves context between sessions. Resume where you left off rather than re-explaining what you were working on. Useful for multi-day projects. Claude remembers the codebase structure, decisions made, and work in progress.

Auto-review workflows: Claude can review code changes, identify issues, apply fixes, and merge, all in a single workflow. For operators managing internal tools or automations, this means code updates can be reviewed and deployed with less manual oversight.

Worktree mode: Work on parallel branches in isolated copies of your repository. Claude creates a separate working copy so you can experiment without affecting your main codebase.

Recurring tasks with /loop (March 2026): Run a prompt on a recurring interval, e.g. /loop 5m check the deploy runs every 5 minutes. Useful for monitoring, status checks, and lightweight automation without leaving Claude Code.

Voice mode (March 2026): Use /voice to interact with Claude Code by voice. Push-to-talk with the spacebar. Supports 20 languages. Useful when you’re thinking through a problem and want to describe it verbally rather than typing.

Effort control (March 2026): The /effort command sets thinking depth: low, medium, or high. Opus 4.6 now defaults to medium effort. Low is fast for simple tasks; high triggers deeper reasoning for complex problems.

1M context support (March 2026): Max, Team, and Enterprise users get 1M token context in Claude Code sessions with Opus 4.6 and Sonnet 4.6, at standard pricing.

Output and model updates (March 2026): Opus 4.6 default output increased to 64k tokens (128k upper bound). /fork renamed to /branch (alias preserved). Opus 4 and 4.1 removed from the first-party API. Users are automatically on Opus 4.6.

Channels (research preview, March 2026): Now supports Slack, Discord, Telegram, and webhooks. Push messages into Claude Code sessions from your team’s chat platform. Research preview, expect changes.

--bare flag (March 2026): Minimal output mode for scripted calls. When you’re using Claude Code in a pipeline or automation script, --bare strips the UI chrome and returns just the result.

--console flag (March 2026): Authenticate via API billing rather than your Claude account. Useful for CI/CD pipelines and server environments where interactive login isn’t practical.

VSCode /remote-control (March 2026): Bridge from VSCode to claude.ai/code. Start a session in your editor and hand it off to the cloud-based Code environment.

Performance improvements (March 2026): ~500ms faster startup and ~18MB memory reduction. Also includes fixes for voice mode issues, concurrent authentication, and Windows PATH handling.

v2.1.90-92 (late March to early April 2026): /powerup interactive tutorial with 18 lessons for learning Claude Code features. /cost command shows per-model token and cost breakdown. Plugin bin/ support for bundling executables with plugins. 60% faster diff computation for large file edits. Bedrock setup wizard for AWS users. Named subagents for organising complex multi-agent workflows. MCP 500K result persistence for keeping large tool outputs across turns.

v2.1.93-101 (April 2026): /team-onboarding command for standardised team setup. OS CA trust for enterprise proxy environments. Bedrock and Mantle integration improvements. Named subagents for organising complex multi-agent workflows.

Source leak incident (31 March 2026): v2.1.88 accidentally included a 59.8 MB source map in the npm package, exposing approximately 512,000 lines of TypeScript including the system prompt and 44 unreleased feature flags. No customer data was exposed. Anthropic described it as “a release packaging issue” and published a corrected version promptly.

I have 500 customer records in a CSV that need to be cleaned: remove duplicates, standardise phone numbers, and split into regional files.

Open the Code tab, describe the task, and Claude writes and runs the script. You get the cleaned files without writing code yourself.

Build me a simple dashboard that reads our sales data from a Google Sheet and displays key metrics.

Claude writes the code, sets up the project, and delivers a working dashboard.


Something wrong or outdated? Let us know →

Get weekly workflows. Subscribe to the newsletter.