Claude Code vs Claude Desktop: Chat, Cowork, Code

Claude Code and Claude Desktop used to be easy to separate: a chat app on one side, a terminal agent on the other. That split no longer matches the product.

As of September 2026, Anthropic's desktop documentation describes one app with three tabs: Chat for conversations, Cowork for Dispatch and longer agentic work, and Code for software development. The Code tab is Claude Code, with a graphical interface around it.

So the useful questions have changed. First, which tab fits the job? Then, if the job is code, do you want the Code tab or Claude Code in a terminal?

Chat: for thinking and drafting

Chat is the conversation surface: questions, drafts, documents you attach, and tools you've connected. MCP servers you define in claude_desktop_config.json are available here, and per Anthropic's docs they also load into local Code tab sessions.

It's the right tab when the output is an answer or a draft you'll judge yourself — explaining an error, talking through an approach, interrogating a document. When the work needs to happen inside a folder on your machine, the other two tabs are built for that.

Cowork: agentic work without a terminal

Anthropic describes Cowork as using "the same agentic architecture that powers Claude Code, with no terminal required" (getting started with Cowork). You give it a goal and it works through the steps across your files and tools. The Cowork product page lists local folders you designate, connected apps such as Google Drive and Slack, and a built-in browser.

The permission model is scoped two ways:

• By folder. Claude can only read and write files in folders you've connected.
• By approval mode. You choose between approving actions manually, automatic approval (Claude reviews each action for safety before it runs), or skipping approvals. Per Anthropic's Cowork safety guidance, it always asks before permanently deleting files, in any mode.

Cowork suits the non-code version of agentic work: a folder of contracts into a comparison table, meeting notes into a weekly summary, a spreadsheet checked against exports. Connecting a dedicated working folder, rather than everything you own, keeps the blast radius small.

Code: Claude Code with a GUI

Before your first message you pick an environment — Local for your machine, Cloud for a session that continues after you close the app, or an SSH connection to a machine you manage — and the project folder Claude works in.

From there, the Code tab does what the old "desktop means chat" framing said it couldn't. It edits files, runs commands in an integrated terminal, runs your tests, and previews your running app in a Browser pane while Claude verifies its own changes. A diff view lets you review and comment on changes, and it can watch the resulting pull request through CI.

What Claude may do without asking depends on the permission mode:

Cloud sessions support Accept edits, Plan and Auto, but not Bypass permissions.

The Code tab vs Claude Code in the terminal

Same engine, same configuration. Per Anthropic's docs, Desktop and the CLI read the same CLAUDE.md files, the same MCP servers configured in ~/.claude.json or .mcp.json, the same hooks, skills and settings, and the same models are available in both. You can move a terminal session into the app by running /desktop (on macOS and x64 Windows, signed in with a Claude subscription).

The differences are about workflow:

Agent teams, where a lead Claude hands tasks to teammates from a shared list, are also CLI-only for now. And if you authenticate with a Console API key, the CLI is the route — Anthropic's Linux install page notes that Desktop doesn't accept one directly.

The rule of thumb: the Code tab when you want to watch, review diffs and see the preview; the CLI when you're scripting it, running it in CI, or already live in a terminal.

Picking, per task

• "Explain this error" → Chat. It's a question.
• "Turn this folder of meeting notes into a weekly summary" → Cowork. Files, several steps, no code.
• "Fix this failing test" → the Code tab or the CLI. There's a pass/fail signal.
• "Add the field end to end and show me it working" → the Code tab, where the preview pane shows the result.
• "Run this check on every pull request" → the CLI, scripted with --print in your pipeline.
• "What should the architecture be?" → Chat. Judgment, no grader.

The pattern: if something can automatically tell you it worked, use Code. If the work is spread across ordinary files and apps, use Cowork. If the answer is a judgment, use Chat.

Cost and access

Claude Code requires a Pro, Max, Team, Enterprise, or Console account; per Anthropic's setup docs, the free Claude.ai plan doesn't include it. Cowork is available on Pro, Max, Team and Enterprise. Some Desktop features are narrower still — computer use and Dispatch need a Pro or Max plan. Check Anthropic's pricing for current details rather than any comparison table, this one included.

The usage shape differs too. A long agentic session in Cowork or Code carries more context per request than a quick chat exchange, so scoping tasks short helps the bill and the output at the same time.

The honest summary

Most people who use more than one tab settle into a rhythm: Chat for thinking, Cowork for file-and-app work, Code for anything with a build or a test. The terminal version is the same Claude Code for people who'd rather script it or never leave the shell.

If you're deciding where to start and you don't have a codebase, start with Chat and try Cowork on one real folder. If you do have one and there's tedious work not getting done, that's the case for the agent — is Claude Code worth it works through that decision, and Claude Code use cases sorts tasks by how reliably they work. For the editor side, Claude Code vs Cursor covers where editors and terminal agents differ.

A different problem from choosing a tab: running an AI setup someone else published without first reproducing their environment. Taku mirrors a working setup into a desktop workspace on your Mac, where you can run it and remix it into your own. The free app library shows what's available to mirror. Taku is in Beta, and the Mac app is available now.

FAQ

What's the difference between Claude Code and Claude Desktop?

Claude Desktop is one app with three tabs: Chat, Cowork and Code. The Code tab is Claude Code with a graphical interface, and Claude Code also runs in the terminal. So the real comparisons are between the tabs, and between the Code tab and the CLI.

Can Claude Desktop edit files and run commands?

Yes, in the Code tab. Pick a project folder and it can edit files, run commands in an integrated terminal, run tests, and preview your app, with a permission mode deciding what it asks before doing. Cowork also reads and writes files, limited to folders you've connected.

What's the difference between Cowork and the Code tab?

Cowork is for longer, multi-step work across files, apps and the web without a terminal — reports, document review, spreadsheets. The Code tab is for software development, with a diff view, an integrated terminal, and an app preview.

Is the terminal version still worth using?

Yes if you script Claude Code with --print or the Agent SDK, run it in CI, authenticate with a Console API key, or want agent teams — Anthropic's docs list those on the CLI side. Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry work directly in the CLI; Desktop reaches them only through a separate enterprise deployment. Both read the same CLAUDE.md, MCP, hook and skill configuration, so moving between them is cheap.

Do they cost the same?

Claude Code requires a Pro, Max, Team, Enterprise, or Console account, and the free plan doesn't include it. Cowork needs a paid plan too. Check Anthropic's pricing page for current details.

Can they all use MCP servers?

Yes. Servers in claude_desktop_config.json load into Chat and local Code tab sessions, and servers in ~/.claude.json or .mcp.json work in both the Code tab and the CLI. Best MCP clients compares hosts more broadly.

Key points

• Claude Desktop is one app with Chat, Cowork and Code tabs, and the Code tab is Claude Code.
• Cowork reads and writes only in folders you connect, and always asks before permanently deleting files.
• The Code tab edits files, runs commands and tests, and previews your app, under a permission mode you choose.
• The CLI shares configuration with Desktop and adds scripting, CI, API-key sign-in and dontAsk.
• Pick by the work: Chat to think, Cowork for files and apps, Code for anything a test can check.