Claude Cowork: The Definitive Setup Guide

What Cowork Actually Is Cowork is an agentic desktop tool built into the Claude Desktop app. It gives Claude direct read/write access to folders on your computer, the ability to execute multi-step tasks autonomously, and the capacity to coordinate parallel sub-agents, all without a terminal or command line. It launched January 12, 2026. Availability expanded to Pro subscribers on January 16, Team and Enterprise on January 23, and Windows (x64) on February 10. On February 24, Anthropic shipped a major enterprise update with new plugins, connectors, and admin controls. Key distinction from regular Claude chat: Chat is prompt-response. Cowork is task delegation. You describe an outcome, Claude makes a plan, breaks it into subtasks, executes in a sandboxed VM on your machine, and delivers finished files to your folder. You can step away and come back to completed work. Cowork is built on the same agentic architecture as Claude Code (Anthropic's terminal-based coding tool). Boris Cherny reportedly built Cowork in ~10 days using Claude Code itself. Still a research preview. Anthropic is explicit: agent safety is under active development. Treat it accordingly. Requirements Platform macOS (universal) or Windows (x64 only). No arm64 Windows. No mobile. No web. Subscription Any paid plan: Pro ($20/mo), Max 5x ($100/mo), Max 20x ($200/mo), Team ($25+/seat/mo), Enterprise (custom) Desktop App Latest version from claude.com/download Internet Active connection required throughout the session App must stay open Closing the desktop app kills the session. Sleep is fine, quitting is not. Step 1: Install and Access Cowork Download the Claude Desktop app from claude.com/download. Sign in with your paid account. Open the app. Find the mode selector at the top, click the "Cowork" tab to switch from Chat to Tasks mode. You're in Cowork. Step 2: Set Up Your Working Folder Cowork operates on a folder-permission model. You grant Claude access to a specific directory, and it can read, edit, and create files within it. The dedicated workspace approach Create a dedicated folder for Cowork rather than pointing it at your entire home directory or Documents folder. This is a safety best practice confirmed by Anthropic's own safety documentation. ~/Claude-Workspace/ ├── context/ # Your standing context files ├── projects/ # Active project folders │ ├── client-a/ │ └── client-b/ └── outputs/ # Where Claude delivers finished work Why this matters: Cowork runs in a VM but has real read/write/delete access to any folder you share. Anthropic's safety guide explicitly warns: "Claude can take potentially destructive actions (such as deleting local files) if it's instructed to." A dedicated workspace limits blast radius. Backup first. Before your first real task, back up anything in the folders you plan to share. cp -R ~/Claude-Workspace/ ~/Claude-Workspace-Backup/ is cheap insurance. Cloud sync for cross-device access Cowork is desktop-only with no built-in sync. If you work across machines, put your workspace in a cloud-synced folder (iCloud Drive, Dropbox, OneDrive) so at least your files are consistent. This won't sync Cowork sessions, just the files. Step 3: Write Your Context Files This is the highest-leverage setup step. The quality of Cowork's output is directly proportional to the quality of context you provide in files. Create these in your context/ folder as .md (Markdown) files. Markdown is the most token-efficient format for Claude to read. about-me.md brand-voice.md working-preferences.md The compounding effect: These files get better over time. After every session where Claude's output missed the mark, update the relevant context file. You can also ask Claude Desktop to help you write these three files based on all previous discussions you got together. Step 4: Set Global and Folder Instructions Instructions are standing directives that load automatically at the start of every Cowork session. Global Instructions: Go to Settings > Cowork in Claude Desktop. Click "Edit" next to Global Instructions. Write your core preferences and save. What to put here (distilled from your context files): Keep this concise. Global instructions load every session, consuming context window. Folder Instructions: Folder-specific instructions activate when you select that folder in Cowork. Use these for project or client-specific context. Example: Your projects/client-a/ folder might have instructions like: Claude can also update folder instructions during a session based on what it learns. Step 5: Install Plugins Plugins bundle skills, slash commands, connectors, and sub-agents into role-specific packages. How to install: In Cowork, click the "Customise" menu in the left sidebar. Click "Browse plugins" to see available options. Click Install on your chosen plugin. Or browse at claude.com/plugins. What's available (as of February 27, 2026) Anthropic open-sourced 11 plugins on January 30 and shipped additional plugins on February 24. The open-source repo is at github.com/anthropics/knowledge-work-plugins. Core plugins (shipped January 30): Productivity, Marketing, Sales, Finance, Data Analysis, Legal, Product Management, Customer Support, Enterprise Search, Biology Research, Plugin Management (for building custom plugins). February 24 additions: HR, Engineering, Design, Operations, Financial Analysis, Investment Banking, Equity Research, Private Equity, Wealth Management. Plus a partner-built Brand Voice plugin by Tribe AI. Each plugin is a folder containing: .claude-plugin/plugin.json - Manifest .mcp.json - Tool connections commands/ - Slash commands you trigger explicitly (e.g., /sales:call-prep) skills/ - Domain knowledge Claude draws on automatically Plugins are just markdown files. You can customise any plugin or build your own with no code. Which to install first: Start with Productivity (useful regardless of role) and one role-specific plugin that matches your job function. Type / in any Cowork chat to see available slash commands from installed plugins. Customising plugins: After installing, click "Customise" on the plugin to work with Claude to adjust skills, commands, and connectors to match your workflow. The default plugins are starting points, the real value comes from adding your company's specific context, terminology, and processes. Step 6: Connect Your Tools Connectors link Claude to external services via MCP (Model Context Protocol). Once connected, Claude can pull live data from these tools during Cowork sessions. How to connect: Go to Settings > Connectors in Claude Desktop (or visit claude.ai/settings/connectors). Browse available integrations. Click a connector and authenticate. Available connectors (as of February 24, 2026): Verified from Anthropic's announcements: Google Drive, Gmail, Google Calendar, Slack, Notion, Asana, Linear, Jira, Monday, ClickUp, Figma, Amplitude, Pendo, Intercom, HubSpot, Close, Clay, ZoomInfo, Fireflies, Microsoft 365, Snowflake, Databricks, BigQuery, Hex, Box, Egnyte, DocuSign, FactSet, MSCI, Harvey, Apollo, Outreach, SimilarWeb, LegalZoom, WordPress, Canva, Ahrefs, Klaviyo, Guru, Benchling, and others. First connector to set up: Connect whichever tool you use most. Slack, Google Drive, or Notion are the highest-leverage starting points for most knowledge workers. Step 7: Run Your First Task Verification prompt: Start a new Cowork session with your workspace folder selected. Type: If the output accurately reflects your context files and global instructions, your setup is working. If something's off, fix it now. First real task: Pick something you already know how to do well, you need to be able to evaluate whether the output is correct. Watch what happens. Answer the questions. Iterate. The best early tasks are things like: Organising a messy folder by type and date Creating a formatted report from scattered notes Building a spreadsheet from receipt images Drafting a document based on existing templates in your folder Scheduled tasks: Cowork now supports recurring tasks. Type /schedule in any Cowork task, or click "Scheduled" in the left sidebar. Tasks only run while your computer is awake and the app is open. Safety: What You Need to Know This section matters more than any productivity tip. Cowork has real power over your files. Anthropic's safety documentation is clear about the risks. Verified safety features: Deletion protection: Claude requires explicit permission before permanently deleting files. You'll see a prompt and must click "Allow." VM isolation: Cowork runs in a sandboxed virtual machine (Apple's Virtualization Framework on macOS). Code executes in an isolated space, but Claude can make real changes to files you've shared. Model training against prompt injection: Anthropic uses reinforcement learning to train Claude to refuse malicious instructions and content classifiers to flag potential injections. Real risks to manage: Prompt injection. If Claude reads a malicious document or website, hidden instructions could alter its behavior. Limit Claude in Chrome access to trusted sites. Web content is the primary injection vector. File destruction. Claude can modify or delete files in any folder you share. Use a dedicated workspace. Keep backups. Include "never delete files without my explicit confirmation" in your global instructions. Scope creep. Watch for unexpected patterns: Claude accessing files you didn't mention, or task scope expanding beyond what you asked for. If something feels off, stop the task immediately. No audit logging. Cowork activity is not captured in audit logs, Compliance API, or data exports. Do not use Cowork for regulated workloads. Session persistence. The app must stay open. Closing it kills the task mid-execution with no recovery. Safety defaults to set: Add these to your global instructions: Cross-App Workflows Cowork can now pass context between Excel and PowerPoint add-ins. Claude can analyse data in Excel and move a chart directly into a presentation without you switching apps. Requirements: Mac users on Max, Team, or Enterprise plans. Both the Claude in Excel and Claude in PowerPoint add-ins must be installed. Windows support isn't available yet for this feature. Caution: Be aware that data from one application may flow into another during a Cowork session. Avoid working with sensitive information in these add-ins while Cowork is active. This guide was compiled from Anthropic's official Help Center documentation, the Cowork blog post, the knowledge-work-plugins GitHub repository, CNBC reporting, VentureBeat coverage, and multiple independent reviews like @heynavtoor's article, then verified against the source material for accuracy. Thanks for reading!