What Claude Cowork is
Anthropic's help page says Cowork brings Claude Code's agentic capabilities to knowledge work beyond coding, with no terminal required. You give it a goal, and it works across your files, your browser and the apps you allow. The work runs on Anthropic's servers in an isolated environment and reaches your computer through the Claude Desktop app.
Older tutorials show a Cowork tab and tasks that run on your own computer. Since 6 October 2026, new Cowork tasks on Pro, Max and Ultra run in the cloud instead. Anthropic is also folding Cowork into the main app. Its product page says Claude Cowork is now just Claude, and that change is still rolling out.
Computer use lets Claude click, type and move through desktop apps on your screen when no better route exists. Anthropic launched it on 23 March 2026 as a research preview. Its help page now calls it a beta, while the product page still says research preview.
Private plugin marketplaces
Admins can create private plugin marketplaces for their organization.
Source: Anthropic: Claude Cowork (read November 8, 2026)
Computer use
Claude can use your apps and screen, in research preview for Pro, Max and Ultra. It pairs with Dispatch, released the week before, which takes tasks from your phone.
Source: Anthropic: Put Claude to work on your computer (read November 8, 2026)
Built-in browser
When a task needs a site, Claude opens a browser in the Cowork side panel.
Source: Anthropic: Claude Cowork (read November 8, 2026)
New tasks run in the cloud
New Cowork tasks on Pro, Max and Ultra run on Anthropic's servers.
Source: Anthropic: Use Claude Cowork on web, desktop and mobile (read November 8, 2026)
What you need first
Cowork needs a paid plan. Computer use is narrower: three plans, and it always runs through the desktop app. Skills are the one part that works on the Free plan, as long as code execution is on.
| Part | Plans | Where it works |
|---|---|---|
| Cowork | Pro, Max, Ultra, Team and Enterprise | Claude Desktop on macOS, Windows and Linux (beta), plus the web and mobile apps. Local files, browser use and computer use go through the desktop app. |
| Plugins | Every paid plan | Chat, Cowork and Claude Code. Hooks and sub-agents run only in Cowork and Claude Code. |
| Skills | Every plan, Free included, with code execution on | Chat, Cowork and Claude Code |
| Computer use | Pro, Max and Ultra. Not Team or Enterprise yet. | Cowork and Claude Code in Claude Desktop on macOS and Windows. Not on Linux. |
Step 1: install Claude Desktop and make a working folder
- Open claude.com/download and pick your system: macOS, Windows, Windows (arm 64) or the Microsoft Store. Linux is in beta for Ubuntu and Debian, but computer use does not run there yet.
- Install the app and sign in with an account on Pro, Max, Ultra, Team or Enterprise. Cowork does not run on the Free plan.
- Make a new folder for Claude, such as
Claude-work, and copy in only the files your first task needs. Anthropic suggests a dedicated working folder over broad access, and backups of anything important. - Open
Settings>Capabilitiesand check thatCode execution and file creationis on. The newer Settings layout calls itCreate and edit files. Skills need it, and they show greyed out without it.
Step 2: start a Cowork task and pick a permission mode
On the older layout, select Cowork in the bottom left corner of the message box, then describe the task. On the new layout there is no switch: you ask, and Claude uses Cowork when the task needs it.
Claude reads and writes only the folders you connect, through the desktop app and only while it is open. Anthropic's pages do not name the control for adding a folder, so use the one your version shows and pick Claude-work. Then choose a permission mode in the message box.
- Manually approve, once called Ask before acting. Claude pauses before actions, and you choose
AlloworDeny. Use it for first tasks, sensitive files and anything hard to undo. - Automatically approve. Claude keeps working, screens each action and blocks unsafe ones. If it keeps hitting blocks, it goes back to asking. It is the default on Pro, Max, Ultra and Team.
- Skip all approvals, once called Act without asking. Nothing checks the actions. Leave it alone until you trust a task from start to finish.
In every mode, Claude asks before it permanently deletes a file. Enterprise accounts start on Manually approve, and their admins can turn the automatic mode off.
Step 3: install a plugin
A plugin bundles skills, connectors and sub-agents into one install, so it is the quickest way to give Cowork a working method for your job. Plugins are saved to your account, not your computer, and follow you into chat and Claude Code.
- In Claude Desktop, open
Customizein the left sidebar. If your app still has a separateCoworktab, open that tab first. - Open the
Pluginstab, then theDiscovertab. - Pick a plugin and click
Add. Anthropic's Knowledge Work marketplace is there by default. - To add another source, click
Addon thePluginspage, thenAdd marketplace. ChooseBrowse Anthropic sources, orAdd from a repositoryand enter a GitHubowner/repo. - Type
/or click+in the message box to see the skills your plugins added. In Cowork you can also run a plugin command as/plugin-name:command. - To fit a plugin to how you work, open it and click
Customizein the upper right corner. A new Cowork task opens. ClickLet's goand tell Claude what to change.
Install plugins only from sources you trust. A plugin's local MCP server runs on your computer with the same permissions as any other program you run.
Step 4: write a skill that tells Cowork how to do your task
A skill is a folder with a SKILL.md file in it: a name, a description and the steps Claude should follow. Claude reads only the description before it decides to load the skill, so that line has to say when to use it.
The name takes lower case letters, numbers and hyphens, up to 64 characters, and must match the folder name. The description can run to 1,024 characters. Keep the whole file under 500 lines.
The example below gives Cowork a daily job and tells it which route to take on your computer. It looks in the connected folder first and asks before it opens an app. Swap the app and the folder for your own.
---
name: meeting-notes-summary
description: Turns today's meeting notes into a one-page summary in the Claude-work folder. Use when the user asks to summarize, tidy or file meeting notes, or asks for today's meeting summary.
---
# Meeting notes summary
1. Look for today's notes in the connected Claude-work folder first.
2. If they are not there, ask the user before you open the Notes app.
Read today's notes only. Do not edit, move or delete any note in the app.
3. Write summary-YYYY-MM-DD.md in the Claude-work folder with three parts:
Decisions, Action items (who and by when), Open questions.
4. Do not send, share or post the summary. Tell the user where the file is.The skill and its wording are this lesson's own example, not Anthropic's. The rules for name and the description are from Anthropic's guide to custom skills.
Step 5: upload the skill and check that it loads
- Save the file as
meeting-notes-summary/SKILL.md. The folder name and thenameline must match. - Zip the folder itself, not the files inside it. From the folder above it, run
zip -r meeting-notes-summary.zip meeting-notes-summary/, or use your system's compress command on the folder. - In Claude, open
Customize>Skills, click+, then+ Create skill, thenUpload a skill, and pick the ZIP. - Check that the switch next to the skill is on. A skill that is switched off is not available to Claude.
- If you would rather not write the file, ask Claude to draft it with the
skill-creatorskill, then read and edit what it wrote.
Claude picks a skill when your request matches its description, so you rarely call it by name. If it does not load, make the description say more plainly when to use it, or name the skill in your request.
Step 6: turn on computer use
Claude uses the most precise tool first. A connector, such as Gmail, Google Drive, Microsoft 365 or Slack, comes first. Next is a browser, either the one built into Claude Desktop or your own Chrome through Claude in Chrome.
Clicking and typing on your screen comes last, for apps with no other way in. Claude asks permission before it uses each app.
- Update Claude Desktop to the latest version.
- Open
Settings>Capabilities. If there is noComputer usesection there, look underSettings>System, inThis computer. - Turn on
Enable computer use. - On macOS 15 or later, Claude works in the background by default. To let it take over the screen, set
How Claude uses the apps you allowtoFull controlin the same section. - Add the apps Claude must never touch to the blocklist. Requests for a blocked app are denied automatically, and investment, trading and cryptocurrency apps are blocked by default.
Keep the computer awake and Claude Desktop open while a task uses your screen. A cloud task keeps running if you close the app, but it can no longer reach your files or apps.
Step 7: run a first task and check the result
Start with a task that only reads and writes inside your folder. This one calls the skill from step 4 by name, so you can see it load.
Use my meeting-notes-summary skill. Summarize today's meeting notes into the Claude-work folder. If you need the Notes app, ask me first. Do not send, share or delete anything.- Switch the mode to
Manually approveand send the task. - Read the plan Claude shows before it starts. If it names a tool or a folder you did not expect, correct it in the chat.
- When Claude asks to open an app, check that it is the one you named, then choose
Allow. - Watch the first steps, and stop the task if anything looks off.
- Open the summary file and compare each decision and action item with your notes. The file is the result, not Claude's message about it.
- Once the task has run clean a few times, type
/schedulein a Cowork task to repeat it. A scheduled run reaches your folder only while your computer is on and Claude Desktop is open.
What Anthropic tells you to keep out of reach
Computer use has no sandbox between Claude and your applications. Claude takes screenshots as it works, so close files and apps with private information first. Anthropic's Action review scans for signs of prompt injection, and the help page calls that capability early.
The help page gives a case in point. Claude clicks a link in an email app, and Chrome opens. Anthropic can keep Claude from seeing the Chrome window, but it cannot stop the link from opening.
You stay responsible for what Claude does for you, including messages it sends, purchases and what scheduled tasks do. Anthropic strongly advises against computer use for these jobs:
- Managing financial accounts or investments.
- Handling legal documents or contracts.
- Processing medical or health information.
- Working in apps that hold other people's personal information.
Common mistakes
- Zipping the files instead of the folder. A
SKILL.mdat the root of the ZIP is not read as a skill. - Naming the folder differently from the
nameline, which makes the upload fail. - A description that says what the skill does but not when to use it, so Claude cannot tell when to load it.
- Uploading a skill from someone you do not know without reading its scripts. Anthropic warns of data leaks from bundled code and says to keep keys and passwords out of skills.
- Starting in
Skip all approvalson a task you have never watched. - Connecting your whole home folder. Claude can read and write everything you connect.
- Looking for computer use on a Team or Enterprise plan, or on Linux. It is not there yet.
Try it yourself
- Make the
Claude-workfolder, put two short text files of meeting notes in it and run the task from step 7. - Cut the skill's description down to what it does, upload it again and see whether Claude still loads it without being told.
- Add one app to the blocklist, ask Claude to use it and read how Claude answers.
- Install one plugin, type
/and list the skills it added.
What to learn next
This is the last lesson of the course. To write descriptions that load at the right moment, take Skill description in the Agent Skills course.
For the sandboxes and permission rules that fit around any agent, read AI agent guardrails.
Check yourself
Which ZIP uploads as a skill?
Show answer
B. A ZIP holding the folder meeting-notes-summary, with SKILL.md inside that folder Anthropic's guide says Claude looks for the skill's folder with SKILL.md inside it, and a SKILL.md at the root of the ZIP is not read as a skill.
In what order does Claude reach for its tools in Cowork?
Show answer
B. A connector, then a browser, then clicking and typing on your screen The help page says Claude uses the most precise tool first: a connector, then a browser, then your screen.
Which permission mode fits a first task you have never watched?
Show answer
A. Manually approve In Manually approve, Claude pauses before actions so you can choose Allow or Deny.
Who can turn on computer use today?
Show answer
B. Pro, Max and Ultra, in Claude Desktop on macOS or Windows The help page says Team and Enterprise plans do not have access yet, and the download page says it is not on Linux.
Frequently asked questions
Is Claude Cowork free?
No. Cowork needs a paid plan, from Pro up to Enterprise. Skills on their own work on the Free plan in chat, with code execution turned on.
Can Claude Cowork use my computer on Windows?
Yes. Computer use is in beta in Claude Desktop for macOS and Windows on Pro, Max and Ultra. It does not run on Linux, and Team and Enterprise plans do not have it yet.
What is the difference between a skill and a plugin?
A skill is one folder that teaches Claude one way of working. A plugin is a package from a marketplace that can hold several skills plus connectors and sub-agents, added in one step.
Does Cowork keep working when I close my laptop?
A cloud task keeps running, but it cannot reach your local files or apps until Claude Desktop is open again. Scheduled tasks run in the cloud even when your computer is off.
Will a skill I upload also work in Claude Code?
Yes, in Claude Code 2.1.273 or later signed in with the same Claude account. The sync runs one way, and it does not happen when Claude Code uses an API key or a cloud provider such as Amazon Bedrock.
Is Cowork the same as Claude Code?
No. Anthropic says Cowork uses the same agentic architecture as Claude Code with no terminal required, and aims it at knowledge work beyond coding.
Sources
- Anthropic: Getting started with Claude Cowork https://support.claude.com/en/articles/13345190-getting-started-with-cowork
- Anthropic: Use Claude Cowork on web, desktop and mobile https://support.claude.com/en/articles/15520349-use-claude-cowork-on-web-desktop-and-mobile
- Anthropic: Let Claude use your computer in Cowork https://support.claude.com/en/articles/14128542-let-claude-use-your-computer-in-cowork
- Anthropic: Use Claude Cowork safely https://support.claude.com/en/articles/13364135-use-cowork-safely
- Anthropic: Use plugins in Claude https://support.claude.com/en/articles/13837440-use-plugins-in-claude
- Anthropic: Use skills in Claude https://support.claude.com/en/articles/12512180-use-skills-in-claude
- Anthropic: Create custom skills https://claude.com/docs/skills/how-to
- Anthropic: Put Claude to work on your computer https://claude.com/blog/dispatch-and-computer-use
- Anthropic: Claude Cowork https://claude.com/product/cowork
- Anthropic: Download Claude https://claude.com/download
