What is Claude Cowork? Pricing, plugins and Cowork vs Claude Code (2026)

Claude Cowork explained: what it does, which plans include it (Pro from $20, not Free), Windows, Mac, Linux, web and mobile, plugins and connectors, scheduled tasks, how it compares with Claude Code, OpenClaw and Perplexity Computer, and what admins should check first.

Walma Engineering·Updated 21 October 2026·12 min read

Claude Cowork is the part of Claude that does work instead of answering questions. You give it an outcome, such as "reconcile these three exports and flag the differences" or "brief me on these five accounts before Thursday", and it plans the task, runs it step by step and hands back files.

It has also changed more than almost any Anthropic product this year. It launched as a Mac-only preview in January, reached every paid plan in April, came to the web and phones in July, and since 6 October runs in the cloud by default. Much of what is written about it describes a version that no longer exists. This guide describes the current one.

Facts are from Anthropic's product pages, pricing page and help center, checked on 21 October 2026. Where only third-party sources exist, we say so.

Claude Cowork at a glance

What it isAn agent for knowledge work: research, analysis, documents
Who makes itAnthropic, same agent architecture as Claude Code
PlansPro, Max, Team, Enterprise. Not Free
PriceNo extra charge; from $20/month with Pro
PlatformsmacOS, Windows, Linux (beta), web, iOS, Android
Where tasks runIn the cloud by default (Pro, Max) since 6 Oct 2026
Extends withPlugins, skills, connectors (MCP), scheduled tasks
StatusGenerally available since 9 April 2026

What Claude Cowork actually does

A normal Claude chat answers one prompt at a time. Cowork takes a goal and carries it out: it writes a plan, splits it into subtasks, runs code to read and transform your files, and delivers a result you can preview and download. You can watch, steer it mid-task, or leave and check in from another device.

Anthropic is folding the two together. Chat and Cowork now share one message box on Pro and Max, and Claude decides whether your message needs a quick answer or a task. Anthropic's own pages have started to say "Claude Cowork is now just Claude". On Enterprise the merged view is a beta switch an admin turns on.

The use cases Anthropic lists are telling, because none of them are about writing:

  • Recurring reports that pull from the same sources every week
  • Building spreadsheets and reconciling financial exports
  • First-pass contract review against your own playbook
  • Analysing sales calls for objections, and account research before a call
  • Going through a pile of notes, tickets or survey answers and summarising them

Anthropic says about half of Cowork usage is "the work around the work": the collecting, cleaning and formatting that sits between an employee and the actual decision.

Claude Cowork pricing: is it free?

No. Cowork is included in every paid plan and in none of the free ones.

PlanPriceCowork
Free$0Not included
Pro$20/month ($17 annually)Included
Max 5x / 20x$100 / $200 a monthIncluded, 5x or 20x Pro's usage
Team$25 or $125 per seat ($20 or $100 annually)Included, on by default
Enterprise$20 per seat annually + usage at API ratesIncluded, admin controls

There is no separate Cowork fee, but it is not free inside a plan either. Every surface (chat, desktop, mobile, Claude Code and Cowork) draws from the same rolling five-hour and weekly limits, and Anthropic warns that Cowork uses them faster than chat because a task runs many steps. The "Auto" permission mode uses more again, since every action is screened. When you run out you wait for the reset or buy extra usage at API rates.

For the full plan comparison, see Claude plans compared.

Platforms: Mac, Windows, Linux, web and mobile

PlatformStatus
macOSAll paid plans, since January 2026
WindowsAll paid plans, since 10 February 2026 (latest desktop app)
LinuxDesktop beta since 30 June 2026: Ubuntu 22.04+, Debian 12+, x86_64 and arm64; no computer use
Web (claude.ai)Pro, Max and Team (Team in beta); Enterprise when an admin enables it
iOS and AndroidSame as web

Some things still need the desktop app open, wherever the task was started: local files, local connectors (including plugins that run a local MCP server), browser use and computer use. If a cloud task needs one of those and your computer is asleep, it waits.

Two related features are worth knowing by name:

  • Dispatch let you message Claude from your phone and have it work on your desktop. It is a limited beta, and Anthropic has closed it to new users; cloud Cowork does the same job without the desktop.
  • The built-in browser (beta since 26 August 2026, Pro, Max and Team) lets Cowork browse inside the desktop app. Claude in Chrome is the older route, through a Chrome side panel.

How to use Claude Cowork

  1. Open Claude on the web, desktop or phone and pick Cowork (or just describe a task, if your account has the merged view).
  2. Connect what it needs. Folders: Claude can only reach folders you connect. Connectors: add them under Customize in the sidebar, for example Google Drive, Slack or your CRM.
  3. Describe the outcome, not the steps. "Compare Q3 spend by supplier against budget and draft a one-page summary for the CFO" works better than a list of instructions.
  4. Review the plan before it runs. This is where you catch a wrong file or a wrong assumption.
  5. Choose a permission mode. Manual (the default) asks before each action. Auto screens each action and blocks unsafe ones. Skip runs with no pauses and no checks; keep it for throwaway folders.
  6. Download or keep the result. Deleting the task removes it from Anthropic's storage within 30 days.

Scheduled tasks are the feature most people discover late. Type /schedule or open Scheduled in the sidebar, and a task such as a Monday pipeline report runs in the cloud on its own, with no device switched on.

Global and folder instructions work like a project brief: tell Claude once how your team formats reports or which folder holds the source of truth, and every task picks it up.

Cowork plugins, skills and connectors

A plugin is a bundle for one kind of work: skills (instructions), connectors (MCP servers), slash commands and sub-agents, packaged as files. Anthropic publishes open-source plugins for:

  • Productivity and enterprise search
  • Sales, marketing and customer support
  • Finance (extended in February with Excel and PowerPoint work) and data
  • Legal, including contract review against a playbook
  • Product management and biology research
  • Plugin create/customize, for building your own

Install one under Customize → Plugins, or upload a plugin file. Plugins you save also work in chat and in Claude Code, so a team writes them once.

Skills are the same format as everywhere else in Claude: a folder with a SKILL.md file. The ones worth starting with are in our list of the best Claude skills; document skills for Word, Excel, PowerPoint and PDF matter most in Cowork. You can even create a skill on Mac by recording yourself doing a task, see Claude skill creator.

Connectors are MCP servers. Cowork can use remote connectors from the directory and local MCP servers on your machine, which is where most of the risk sits (see below). Our guide to Claude connectors covers how they work, and the MCP directory lists the servers for common business tools.

On Team and Enterprise, an owner can run a private plugin marketplace under Organization settings → Plugins, from a ZIP or a private GitHub repo that syncs automatically. That is how a company gives everyone the same reviewed legal or finance plugin instead of 40 people installing their own.

Claude Cowork vs Claude Code

Claude CoworkClaude Code
Built forResearch, analysis, documentsSoftware engineering
Works inFolders, connectors, browserA code repository, terminal, IDE
OutputFiles, reports, answersCode, commits, pull requests
InterfaceClaude app, web, phoneTerminal, IDE, desktop, web
RunsCloud by default, local when neededLocal by default, cloud optional
UsageSame poolSame pool

The rule of thumb: if the result is code, use Claude Code. If it is a document or a decision, use Cowork. Developers end up using both, and plugins carry across. Work that must never leave the machine belongs in Claude Code in the desktop app, which Anthropic itself points to for that case.

Cowork vs OpenClaw and Perplexity Computer

These comparisons come from third-party sources; neither vendor publishes one.

Claude CoworkOpenClawPerplexity Computer
TypeManaged productOpen-source, self-hostedManaged cloud agent
ModelsClaude onlyAny (Claude, GPT, Gemini, local)About 19, orchestrated
CostFrom $20/month planFree software + your API costsPerplexity Max $200/month, credit-based
SetupInstall the appNode.js, API keys, configurationSign up
Main riskConnectors and unattended tasksFull system access, unvetted skillsCredits run out on heavy use

OpenClaw suits someone who wants to run an agent on their own server with any model and is comfortable owning its security. Cowork suits a team that wants a managed agent tied to the plans it already pays for.

Security and admin: what to check before a rollout

Anthropic's own safety article is unusually direct about the risks. The ones that matter for a company:

  • Prompt injection. A document or web page can contain instructions. Anthropic says the chance of an attack is "still non-zero".
  • Write and delete actions. Cowork can change and delete files; permanent deletion needs explicit permission.
  • Computer use runs without the permission checks that gate the other tools.
  • Egress rules do not cover web fetch, web search, MCP or Chrome.
  • Local MCP servers run with the user's own permissions, and data can move between connected apps.
  • Scheduled tasks run unattended, so a bad instruction repeats every week.

What admins can control on Team and Enterprise:

ControlDefault
Cowork on/off for the organizationOn
Run Cowork in the cloudOn for Team, off for Enterprise
Built-in browserOn (Enterprise since 10 Sep 2026, unless turned off)
"Always allow" for connector toolsOff
Per-tool connector rules (e.g. read yes, write no)Enterprise
Role-based access via SCIM groups, group spend limitsEnterprise
OpenTelemetry events for every tool call, file read and change, approvalTeam and Enterprise

Two gaps to know. Local session history stays on the user's machine, outside Anthropic's retention policy, and admins cannot delete it. And the Compliance API covers Cowork sessions, but local sessions are only retrievable for the last 30 days. Anthropic also says OpenTelemetry "doesn't replace audit logging for compliance purposes."

The checklist we use: start with Manual mode, allow read-only connector tools first, keep scheduled tasks to a named owner, review plugins centrally instead of letting everyone install their own, and stream OpenTelemetry to whatever already watches your other systems. The MCP security guide goes deeper on connectors.

Data and the EU

Cowork's documentation says nothing about EU data residency, and as of October 2026 Claude's own plans offer no EU processing option. With cloud execution now the default on Pro and Max, files a task needs are copied to Anthropic's environment for the length of the session. For most personal use that is fine. For a European company handling customer, employee or financial data, it is the question legal will ask first.

The usual answers are Claude through a cloud provider's EU region (Anthropic lists Amazon Bedrock, Google Cloud and Microsoft Foundry as routes to Cowork), or a gateway in your own EU tenant. See Claude in the EU for the options.

Fixing "missing HCS services" on Windows

The error "Missing HCS services: HNS, vmcompute, vfpext" means the Windows virtualization services Cowork's local environment relies on are not running. Anthropic has no article on it; the fixes users report are:

  1. Enable Virtual Machine Platform (and Hyper-V where your edition has it) under Windows Features, then restart fully. This needs admin rights, so on a managed laptop it is an IT ticket.
  2. If it persists, clear the cache under Help → Troubleshoot in Claude Desktop and restart the app.
  3. If the message names vfpext, enable the Containers optional feature as well. DISM and SFC repairs reportedly do not help.

Tasks that run in the cloud do not use the local environment, so they may work even while the error remains.

Claude Cowork for a whole company

Cowork is at its best when everyone has the same reviewed plugins, the same connectors and clear rules about what can be read and written. That is a platform decision more than a personal one. Walma AI Hub gives every employee Claude, GPT and other models through one gateway in your own EU region, with per-team model access, budgets, a full audit log and the same skills and MCP connections for everyone, including the department setups for legal, finance and sales. See how it works, or book a 20-minute walkthrough below.

Frequently asked questions

What is Claude Cowork?+

Claude Cowork is Anthropic's agent for knowledge work. Instead of answering one prompt at a time, you describe an outcome, Claude makes a plan, splits it into steps, runs code where it needs to and hands back finished files such as spreadsheets, decks and reports. It is built on the same agent architecture as Claude Code, but for research, analysis and documents rather than software, and with no terminal.

Is Claude Cowork free?+

No. Cowork is included in every paid Claude plan (Pro, Max, Team and Enterprise) at no extra charge, but not in the Free plan. It draws on the same usage allowance as chat and Claude Code, and Anthropic notes that Cowork uses that allowance faster than chat because tasks run many steps.

How much does Claude Cowork cost?+

The cheapest way to get it is Claude Pro at $20 a month ($17 billed annually). Max costs $100 or $200 a month, Team $25 or $125 per seat a month ($20 or $100 billed annually), and Enterprise $20 per seat a month billed annually plus usage at API rates. There is no separate Cowork price.

What is the difference between Claude Cowork and Claude Code?+

Claude Code is for software engineering: it works in a code repository, a terminal and an IDE. Cowork is for everything around it: research, reports, spreadsheets, contract review, account research. They share the same agent approach and the same usage allowance, and plugins you save work in both. If the output is code, use Claude Code; if the output is a document or an answer, use Cowork.

Does Claude Cowork work on Windows?+

Yes. Cowork has been on Windows since 10 February 2026, on every paid plan, with the latest Claude desktop app. Anthropic's current documentation lists Windows x64; whether Windows on Arm is supported is not stated. There is also a Linux desktop beta (Ubuntu 22.04+, Debian 12+) and Cowork on the web and in the mobile apps.

Where does Claude Cowork run?+

Since 6 October 2026, Cowork tasks on Pro and Max run in the cloud by default: code runs in an isolated, temporary environment on Anthropic's servers that is deleted when the session ends. Your folders stay on your computer, and Claude fetches a copy of a file only when a task needs it. Local files, local connectors, browser use and computer use still need the desktop app open.

What are Claude Cowork plugins?+

A plugin is a bundle of skills, connectors (MCP servers), slash commands and sub-agents for one kind of work. Anthropic publishes open-source plugins for productivity, enterprise search, sales, finance, data, legal, marketing, customer support, product management and biology research, plus one for building your own. Install them under Customize, then Plugins. On Team and Enterprise, an owner can run a private plugin marketplace for the organization.

Is Claude Cowork safe for company data?+

It can be, with configuration. Anthropic itself names the risks: prompt injection from documents and web pages, write and delete actions, data moving between connected apps, scheduled tasks that run unattended, and computer use without the usual permission checks. On Team and Enterprise, admins can turn Cowork off, keep it local, restrict connector tools to read-only and stream every action to a SIEM via OpenTelemetry.

Can Claude Cowork keep data in the EU?+

Anthropic's Cowork documentation says nothing about EU data residency as of October 2026, and Claude's own plans offer no EU processing option. Companies that need EU processing usually reach Claude through a cloud provider's EU region or a gateway hosted in their own EU tenant.

How do I fix 'missing HCS services' in Claude Cowork on Windows?+

The error means the Windows virtualization services the local Cowork environment needs are not running. Users report fixing it by enabling Virtual Machine Platform (and Hyper-V where available) in Windows Features and restarting, and in the vfpext case by also enabling the Containers optional feature. Anthropic has no official article on it. Tasks that run in the cloud do not use the local environment.

Walma AI Hub

The same tools, in your EU region, under your control

A 20-minute walkthrough with an engineer. We map it to your tools, your MCP servers and your budget model.

About AI Hub