How to use Claude Cowork, step by step, and the open-source system a team moves to

What you need before the first task

Cowork is available on paid Claude plans (Pro, Max, Team, Enterprise), and the free Claude tier is chat only. For local files, browser use, and computer use, the Claude Desktop app for macOS or Windows must be open and connected, because those abilities reach things on your computer even though the session itself runs in the cloud. Both facts come from Anthropic's get-started guide (checked October 2026).

Step by step

1. Open Claude and pick Cowork. On desktop, web, and mobile, chat and Cowork share one home. Find the message box and select "Cowork"; to return to a plain conversation, select "Chat".
2. Describe the outcome you want. Claude analyzes the request, breaks complex work into subtasks, and creates a plan. Say what finished looks like, and let the plan carry the method.
3. Connect a folder, not the whole machine. For local files, Claude reaches your computer through the desktop app, and only for the folders you have connected. If the app is offline, the session cannot reach your computer.
4. Connect the tools. Cowork reads your connected apps and can use MCP extensions and plugins. Every extension widens what the session can reach, so add the ones you have checked and leave the rest out.
5. Review the plan and set the approval mode before it runs. Cowork offers "Manually approve", "Automatically approve", and "Skip all approvals". Match the mode to the stakes, because a task can act on malicious content before you notice.

The pitfalls that cost the most

• A routine approval stops being a check. "Skip all approvals" runs with nothing screening the actions, and "Automatically approve" screens each one itself. Switch to "Manually approve" when a task touches sensitive files, accounts, or sites, or when a mistake is hard to undo.
• Wide folder access reaches wide file access. Anthropic advises a dedicated working folder, because a session can read, write, and permanently delete the files you connect, and any file it opens is processed on Anthropic's servers.
• Scheduled tasks run unwatched. They run in the cloud with no device online, so start with low-risk jobs, keep sensitive data out, and review the run history from the "Scheduled" page.
• Computer use acts on your screen. File operations pass through permission checks and code runs in an isolated environment, but computer use has no sandbox between Claude and your screen, and it takes screenshots to read it. Block the apps you would rather keep private.

Where a single desktop stops and a platform a team owns starts

Cowork keeps its work in your Claude account, so the unit of work is one person. Three signals say the team has outgrown it:

• More than one person needs the same agents and skills, and the setup should not be rebuilt for each desk.
• The company's memory should be shared and owned rather than spread across personal accounts.
• Someone must review what an agent changed before it lands, every time.

At that point the question changes from how to use Cowork to where the work should live. Kortix is the open-source AI Operating System: your agents, their skills, your company memory, and every connector in one git repo you own. Every session runs on its own cloud computer, and session work reaches main through a change request. Merge is default-deny for agents. It reaches 3,000+ apps in a click, plus MCP, OpenAPI, Postman, GraphQL and raw HTTP. Run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free. Try Kortix on the managed cloud to start, or read the docs for the full model.

The Claude Cowork alternatives roundup compares the options, Kortix vs OpenWork covers the desktop route, and Kortix for teams covers roles and approval gates. To move an existing setup, migrating from Claude Cowork is the checklist.