chapter three

3 Using Claude Desktop and Cowork

 

This chapter covers

  • Moving Claude’s workspace onto your machine
  • Working on local files with Cowork
  • Organizing work into Git-backed Projects
  • Connecting live systems with MCP connectors
  • Building live artifacts and previewing computer use

Claude on the web is a very powerful tool, but using it means making a lot of upload-download round trips. Look at the flow of data so far: you uploaded a sketch and downloaded an architecture document. Then you uploaded that architecture document and downloaded the component designs, in my case eight separate documents. If we carry on in the web, the next stage starts with you uploading a stack of documents into a fresh conversation, and downloading whatever comes back. You have to do that because the web interface can’t cross the network boundary and reach into your machine. You’ll become a data courier.

But we won’t do that. It’s time to move away from the web interface, where everything runs in Anthropic’s cloud. The model will (probably) always run in the cloud, but if the tools can access your local machine then they can read input files and write outputs directly. Claude Desktop is the native app for Mac and Windows, and its Cowork feature gives Claude a working sandbox to run tools on your own machine, with permissioned access to your files and local network services.

3.1 Why you need the desktop app

3.2 Installing Claude Desktop and trying Cowork

3.3 Working with local files and folders

3.4 Projects: Organizing chats, tasks, and data

3.5 Accessing external systems with connectors

3.6 Live artifacts and connectors

3.7 Computer use: Letting Claude drive your desktop

3.8 Your turn: Set up your application project

3.9 Summary