stacktr.ee
By · Founder, Stacktree Last updated
comparison

Claude Cowork vs Claude Code

Same agentic engine, two front doors. Claude Code runs in your terminal against a codebase. Claude Cowork runs in the Claude desktop app against the folders, documents and connected services of a working day. Both come with a paid plan, both take the same plugin packages, and both finish with a file that still has to reach somebody.

Get started free

No card. 3 free pages, each live for 7 days.

What is the difference between Claude Cowork and Claude Code?

Claude Code is the terminal agent for people who work in a codebase. Claude Cowork is the desktop agent for people who work in documents, spreadsheets and connected services. Both run multi-step jobs, both take the same plugin packages, and both arrive with a paid Claude plan rather than a separate subscription of their own.

The honest side by side

Criterion Claude Code Claude Cowork
Audience Developers, and anyone at home in a terminal. Operators, founders, consultants, marketers, finance.
Interface Command line, plus IDE integrations. Claude desktop app on macOS and Windows, with web, mobile and a Chrome side panel on some plans.
What it is good at Reading and changing a codebase, running tests, shipping software. Documents, spreadsheets with working formulas, decks, reconciling records across connected tools.
How it reaches your files The working directory you launched it in. The folders you connect, and nothing outside them.
Plugins and skills Yes, plugin packages with skills, connectors, subagents and hooks. Yes, the same format. Anthropic says plugins run in Cowork and Code, not chat.
Custom marketplaces Yes, from a Git repository. Yes, add a GitHub, GitLab or Bitbucket repo by URL.
Approval model Permission prompts per tool, with modes to loosen them. Manual, auto with safety checks, or skip.
Plans Pro, Max, Team, Enterprise. Not on Free. Pro, Max, Team, Enterprise. Not on Free.
Learning curve A terminal, a config file, and MCP by command. A sidebar, a Customize panel, and Install buttons.
Getting the result to a client Neither one sends anything. Both stop at a file. Neither one sends anything. Both stop at a file.

Where the line actually falls

It is tempting to call this a skill split, as though Cowork were the simple one. It is not. Both run the same kind of job: plan, execute over many steps, touch real systems, come back with something finished. The split is what the job ends in.

Claude Code ends in a repository. Its native objects are files under version control, a test suite, a diff, a branch. Everything about the interface assumes you will read the change before it lands. Cowork ends in a deliverable. Its native objects are the folder on your desktop, the spreadsheet with formulas that calculate, the deck, the record updated in a connected service. Everything about the interface assumes you want the outcome rather than the steps.

If you are choosing for a team, choose by the same test. A product team lives in Claude Code. A services business lives in Cowork, which is exactly why Anthropic shipped Claude for Small Business as a Cowork plugin: 43 ready-to-run workflows built for the operating rhythm of a company, not for a codebase.

Plugins are the shared part

The interesting overlap is packaging. Anthropic's documentation states plainly that plugins are available in Cowork and Code, and are not used in chat. A plugin manifest can declare skills, MCP connectors, subagents and hooks, and installing it adds all of them at once. Cowork accepts a Git repository as a marketplace, which means the same repo can serve both tools.

For anyone building on top of Claude, that is the useful news: you write the capability once. What changes is where it lands and who runs it. The full picture of what Cowork exposes is on Claude Cowork plugins and skills.

Pricing: the same subscription either way

Neither tool is bought on its own. Anthropic's pricing page lists Claude Code and Claude Cowork on the paid plans: Pro at $17 a month billed annually or $20 monthly, Max from $100 a month, Team at $20 a seat annually or $25 monthly with premium seats at $100 and $125, and Enterprise at $20 a seat annually plus usage at API rates. The Free plan has neither.

The cost that surprises people is usage rather than price. An agentic job is dozens of model calls. Whichever surface you pick, running standing workflows every week is a different consumption profile from asking questions in chat, and it is worth sizing the plan for the workflows rather than the conversations.

When to use each

Use Claude Code when

  • The work is a codebase, and the output is a change to it.
  • You want to read a diff before anything is real.
  • Your workflow already runs through the terminal: scripts, git, CI, hooks.
  • You are automating something unattended, on a server or in a scheduled job.

Use Claude Cowork when

  • The work ends in a document, a model, a deck or a set of records.
  • The inputs are scattered across a folder and half a dozen connected services.
  • The person running it is not going to install a CLI, and should not have to.
  • You want packaged workflows to run rather than prompts to write, which is what the plugin directory is for.

Where Stacktree fits, on both sides

Both tools stop at the same place: a file, on a disk, with nobody outside your company able to see it. The send step belongs to neither of them, and it is where most of the value leaks out.

From Claude Code, Stacktree is an MCP server. One command, claude mcp add --transport http stacktree https://api.stacktr.ee/mcp, an OAuth sign-in and no API key, and Claude Code publishes the HTML it wrote to a private, unguessable URL in a tool call. When it revises the file it calls update_site and the URL you already sent shows the new version. That path is documented in full on publish HTML from Claude Code.

From Cowork, Stacktree is the delivery step. The workflow produces a proposal, a report or a page; you publish it and send a link the client opens in any browser with no account, put a passcode on it if the contents warrant one, see when it was opened, and replace it in place after the call. Cowork speaks MCP, so Claude can do the publishing itself, or you can drag the file onto the dashboard. The walkthrough is on sending Cowork output to a client.

Same destination, two routes in. That is the point: the link you hand over does not care which front door made the file.

FAQ

Frequent questions

What is the difference between Claude Cowork and Claude Code? +
The interface and the audience, not the underlying idea. Both are agentic: Claude plans a multi-step job and carries it out against real files and real services. Claude Code does it in your terminal, against a codebase, for people who are comfortable there. Cowork does it in the Claude desktop app, against the folders and connected services of a working day, for people who would rather not open a terminal at all.
Should I use Claude Cowork or Claude Code? +
Pick by what the job ends in. If it ends in a commit, a branch or a running application, use Claude Code. If it ends in a document, a spreadsheet, a deck or a set of records updated across several tools, use Cowork. If it ends in a page somebody outside your company has to read, either one can write it and both need somewhere to put it.
Do Claude Code plugins work in Claude Cowork? +
Plugins are a shared format. Anthropic’s documentation states that plugins are available in Cowork and Code and are not used in chat, and a plugin package can declare skills, MCP connectors, subagents and hooks in one manifest. Cowork can also take a Git repository as a marketplace, so a plugin published as a GitHub repo is installable from either side.
Does Claude Cowork cost extra on top of Claude Code? +
No. Both are included with a paid Claude plan rather than sold separately. Anthropic’s pricing page lists Claude Code and Claude Cowork on Pro, Max, Team and Enterprise. The Free plan has neither. What differs is usage: an agentic job of any kind is many model calls, so both draw down a plan faster than chat.
Can Claude Cowork write code? +
It executes code as part of doing a job, and it works in the folders you connect, so it is not walled off from source files. But the surface is built around documents and connected services rather than around a repository, a diff and a test run. If you are shipping software, the terminal tool is the one shaped for the work.
Can Claude Code produce a spreadsheet or a deck? +
It can write any file it can generate with code, so the answer is technically yes and practically awkward for a non-developer. Cowork is the surface that treats a spreadsheet with working formulas or a formatted presentation as a first-class output rather than a scripting exercise.
Is Claude Cowork replacing Claude Code? +
There is no sign of it. Anthropic lists both on the same paid plans and ships plugins that run in both. They are two entry points to the same agentic capability, aimed at two audiences who want very different interfaces.
Can I use both? +
Yes, and many people will, because most jobs have a part that belongs in a repo and a part that belongs in a document. A useful division: Claude Code builds the thing, Cowork assembles the reporting and the client-facing write-up around it. Both can publish to the same place, so the link you hand over does not care which one produced the file.
How do I share what either one produces? +
Publish it as a page and send the link. From Claude Code, connect the Stacktree MCP server with one command and it publishes the HTML it wrote in a tool call. From Cowork, publish the exported file the same way, by MCP or by dragging it onto the dashboard. Either way the recipient opens a private URL in a browser with no account, and you replace the contents in place when the work changes.
Keep reading

Related guides

References

Sources and further reading

Whichever front door made the file, send it as a link.

An MCP call from Claude Code, or a drag and drop from Cowork. The client opens a private page in a browser, with no account.

Sign up free →