Who Gets Claude Cowork and How to Access It

Cowork is a mode inside Claude, not a separate product. It ships with the paid plans: Pro, Max, Team, and Enterprise. There is no waitlist to join, no separate purchase, and nothing extra to install. If you have a paid plan you already have it. You just have not selected it yet.

That is the whole answer. Most people searching for access are not blocked by a gate. They are blocked by looking for a product page that was never built.

What is Claude Cowork, in one paragraph?

Cowork is Anthropic's interface for handing an agent a folder of real files and a real job, then walking away while it works. Instead of pasting context into a chat window, you give it access to your actual documents and it operates on them directly.

That is the compressed version. The full breakdown of how background agents work, why the folder-level model matters, and what changes when an agent runs without you watching lives in what Claude Cowork is and how background agents work. This post covers something different: whether you can get it, what it costs, and what to do the first hour you have it.

Who gets Claude Cowork?

Anyone on a paid Claude plan. That means Pro, Max, Team, and Enterprise. The free plan does not include it, and no amount of clicking will surface it there.

The important structural fact is that Cowork is bundled, not sold. Anthropic did not build a separate SKU for it. It appears inside a product you already pay for, which is why searching for "how to get Claude Cowork" returns so little useful information. There is no getting. There is only selecting.

If you are on the free plan, what you do have is the standard Claude chat experience, which is a genuinely useful place to test whether the agentic workflow fits how you work before you pay for anything. Run a few tasks in chat. Paste in a document, ask for a structural rewrite, watch how it handles ambiguity. If that feels like leverage, the paid tier is where it stops being a conversation and starts being an employee.

If you are on a Team or Enterprise account, some of this is not yours to decide. Owners can turn off web search for Cowork and Chat from Organization settings, and can control Claude in Chrome the same way. Organizations can also require per-task approval for write-capable connector tools, which means your own "always allow" preference may simply not apply. That is worth a message to whoever administers your workspace before you assume a missing capability is a bug.

Is Claude Cowork free?

No. It requires a paid Claude plan.

This is worth separating from the eligibility question because people ask it for a different reason. "Who gets it" is a question about tiers. "Is it free" is usually a question about whether there is a trial, a limited version, or a way in through the back door.

There is not. There is no free Cowork tier, no capped usage version, and no standalone trial. The free path is the Claude chat experience, which is a different surface with different capabilities.

How much does Claude Cowork cost?

It costs whatever your Claude plan costs. There is no separate Cowork price, because it is not separately sold.

The cost that does vary is usage, and this is the part most people miss. Cowork consumes more of your allocation than chatting does, because a multi-step task runs longer, calls more tools, and burns more tokens than a single exchange in a message box. A task that reads forty files and writes one document is not priced like a question.

The sharper point is that it all comes out of one bucket. Your activity across Claude on the web, on desktop, on mobile, and in Claude Code draws from the same usage pool. Heavy Cowork use eats the same budget as everything else you do with Claude. The plan price is only half the cost question, and the half people plan around is the half that does not move.

How do you access Claude Cowork?

There is no Cowork tab, and looking for one is why this question gets asked so often. Chat and Cowork share a single home. You find the message box, select Cowork, and describe the task. Selecting Chat takes you back. That is the entire navigation story.

It runs in more places than most write-ups suggest:

• Claude Desktop for macOS
• Claude Desktop for Windows, which requires the latest version of the app from claude.com/download
• The web, at claude.ai
• Claude Mobile, on the latest iOS and Android apps

Where the work actually happens

This is the part the rest of the internet gets wrong, and it changes how you should think about the tool.

Cowork runs your tasks remotely, in beta, on Anthropic's servers, inside an isolated environment. It is not running on your laptop. Sessions and the files they produce save to your Claude account and follow you between surfaces, so a task you start on desktop is a task you can check on your phone. Close the laptop and the work carries on without you, which is the entire point of a background agent and the thing a chat window cannot do.

The desktop app still matters, but for a narrower reason than "that is where Cowork lives." You need it when a task has to reach something on your own computer: local file access, browser use, and computer use. The session is still remote. The desktop app is the arm it reaches back through.

The sequence the first time:

1. Open Claude on whichever surface you use, signed in to your paid account.
2. Install the desktop app as well if the job needs local files, browser use, or computer use.
3. Find the message box and select Cowork.
4. Grant access to a folder. This is the step that matters, and the one people rush.
5. Describe the job in plain language and let it run.

Step four is where the whole thing lives or dies. You are not opening a chat window. You are handing an agent the keys to a directory. Pick a small one.

Here is a first task that demonstrates the difference between Cowork and a chat window without risking anything real.

First-run prompt for a copied folder

Read every file in this folder. Do not change anything yet.

Then give me three things:

1. A one-line summary of what each file contains.
2. Any file that appears to be an outdated duplicate of another file, with your reasoning.
3. A proposed folder structure that would make this easier to navigate in six months.

Wait for my approval before creating, moving, or renaming anything.

That prompt is deliberately read-only. It shows you how the agent perceives your files before it touches them, which is the single most useful thing to learn in the first ten minutes.

Is Claude Cowork available on Windows?

Yes. Cowork runs on Claude Desktop for Windows, and it needs the latest version of the app, which you can get from claude.com/download. If you installed Claude for Windows some time ago and Cowork is not there, updating is the first thing to try.

The framing worth keeping is that Cowork's platform support follows the Claude app rather than shipping on its own schedule. People search for "when is Cowork coming to Windows" as though it were a separate product with its own roadmap. It is not. Watch the app, not the feature.

If your work is more terminal-shaped than folder-shaped, Claude Code for non-developers is the sibling tool for the same category of job. It asks more of you in the first hour and gives you more control after it.

Is Claude Cowork safe to use?

Safer than the version of this question most people ask, because you choose how much rope the agent gets before it starts. There are three permission modes, and picking one is most of the safety decision.

Manual. Claude pauses and asks for approval before it acts. You choose Allow or Deny on each action. Slowest, and the right default for the first few sessions.

Auto. Claude keeps working without stopping, and reviews each action for safety as it goes, including data exfiltration and prompt injection, blocking anything it judges unsafe. It consumes more usage than the other modes precisely because of that checking. Auto will not approve sensitive actions on its own: granting access to additional folders, deleting files, and creating scheduled tasks still come back to you.

Skip. No pausing and nothing checked automatically. Fastest, and appropriate only when you already know exactly what the task will touch.

Claude also requires explicit permission before permanently deleting any files. A prompt appears and you have to select Allow. That is true regardless of mode.

Auto mode's safety check is a mitigation, not a replacement for judgment. It reduces the blast radius of a bad input. It does not make a bad input safe, which is why the next habit still matters more than the mode you pick.

Never point it at documents you did not write. This is the one people skip. When an agent reads a document, that document's text becomes input the agent may act on. A PDF from an unknown source, a shared doc, a downloaded contract: any of those can contain instructions aimed at the agent rather than at you. This is prompt injection, and it is not theoretical. The defense is boring and it works: agents read your files, not strangers' files, unless you have read them first yourself.

Two habits sit underneath the modes. Copy the folder before the first run, so the sessions where you learn how the agent interprets vague instructions happen on files you can delete. And scope narrow, then widen, because an agent with access to everything will occasionally reason its way into a file you forgot was there.

Beyond that, treat agent output the way you would treat a capable new hire's first week of work. Read what it changed before you ship it. Approve, do not assume. That habit is the backbone of every reliable setup in the AI agent workflows playbook, and it matters more as the agent gets better, not less.

What can Claude Cowork actually do?

Concrete jobs it handles well, rather than capability categories:

• Read a folder of forty client call transcripts and produce one document listing every objection raised, grouped by theme, with the quote attached.
• Take a directory of invoices and build a spreadsheet of vendor, date, amount, and category.
• Audit a content folder for every post that references an outdated product name and produce a fix list with file paths.
• Read a research folder and draft a brief that cites the specific files each claim came from.
• Reconcile two versions of a document set and produce a change log a human can review in five minutes.

The pattern across all five: the work is tedious, the input is a pile of files, and the output is a single artifact you actually read. That is Cowork's shape. Anything where the bottleneck is reading volume rather than making judgment calls is a candidate.

What it is not is a chat replacement. If your task fits in a message box, use the message box. The overhead of granting folder access only pays off when the folder is the point. The background agents breakdown goes deeper on where that line sits and why so many people put agentic tools to work on tasks that never needed them.

Start with one folder

The access question has a shorter answer than the search volume suggests. Paid plan, open Claude, select Cowork. The harder question is what you hand it first, and that is a question about your operation, not about the software.

If you want the full system for figuring out which parts of your work belong to an agent and which stay yours, I walk through it in the free masterclass. No pitch in the first twenty minutes, just the framework.