Claude Cowork is one of the most underused setups in the Claude ecosystem. It turns Claude from a chat window into something closer to a persistent desktop collaborator — one that can see files, run terminal commands, browse the web, and work alongside you across a real session. If you have Claude Max, you already have access to it. Most people set it up wrong or skip it entirely. This guide fixes that.
What Claude Cowork Actually Is
Claude Cowork refers to using Claude Code — Anthropic's official CLI — as your primary working interface instead of the browser chat. It runs in your terminal, connects to your local filesystem, and can execute code, read files, and take multi-step actions without you copying and pasting between windows.
The practical difference is significant. In browser Claude, you paste content in, Claude responds, you paste the output somewhere else. In Claude Code, Claude operates directly inside your project. It reads the files, writes the edits, runs the commands, and shows you what changed. It is a real working session, not a conversation you have to relay manually.
This is not for developers only. If you run a business and you have folders of docs, spreadsheets, email drafts, SOPs, or marketing assets, Claude Code can work across all of it. The setup takes about ten minutes.
What You Need Before You Start
Check these four things before installing anything. Skipping them causes the most common setup failures.
- Claude Max subscription — Claude Code runs on your Claude account, so you need an active Max plan. The free tier does not include API access at the level Claude Code requires for sustained use.
- Node.js installed — Claude Code is a Node package. Open your terminal and run `node --version`. If you see a version number, you are good. If not, install it from nodejs.org before continuing.
- A terminal you are comfortable opening — on Mac, that is Terminal or iTerm2. On Windows, use PowerShell or Windows Terminal. You will live here during setup.
- A project folder to work in — have one directory in mind where you want Claude to operate. This does not have to be code. It can be a folder of documents, a client deliverable folder, anything.
Installing Claude Code
The installation is a single terminal command. Open your terminal and paste this exactly:
npm install -g @anthropic-ai/claude-codeOnce it installs, run the login command to connect it to your Anthropic account:
claudeThe first time you run `claude`, it will open a browser window and ask you to authenticate with your Anthropic account. Log in, approve the connection, and come back to your terminal. You should see a prompt that confirms you are connected. That is the full install. You are now running Claude as a desktop agent.
To verify everything is working, navigate to your project folder and start a session there. Claude will read the contents of that directory and be ready to work inside it.
cd ~/Documents/your-project-folder
claudeYour First Real Working Session
The fastest way to understand what Claude Code changes is to give it a task that would take you twenty minutes manually. Here are three opening prompts that show what it can actually do:
If you want Claude to audit a folder of documents and summarize what is in it:
Read all the files in this folder and give me a one-paragraph summary of each one. Then tell me which ones overlap in topic and which ones are missing a clear call to action.If you want Claude to draft something and save it directly into the project:
Write a follow-up email sequence for a coaching discovery call. Three emails, spaced three days apart. Save each one as a separate .txt file named followup-1.txt, followup-2.txt, followup-3.txt. Keep the tone direct and warm. No hype.If you have an existing doc and want Claude to rewrite or improve it in place:
Read my-sop.md and rewrite the section on onboarding to be clearer and shorter. Show me the before and after, then ask if I want to save the change.Notice what is happening in each of these. Claude is not just generating text you copy somewhere. It is reading your files, writing outputs, and confirming changes. That is the cowork model.
Setting Up Permissions the Right Way
By default, Claude Code will ask for your permission before writing or editing any file. That is a safe default. As you get comfortable, you can adjust which actions it can take automatically and which ones still require a prompt.
There are two ways to configure this. The first is per-session — you tell Claude in the chat what it is allowed to do during that session. The second is through a settings file that persists across sessions.
For most people getting started, keep the default approval prompts on. As you understand what Claude is doing, you will develop a clear sense of which actions are safe to automate. The goal is not to remove all friction — it is to put yourself in control of where friction lives.
Never grant broad write permissions to folders that contain sensitive data, credentials, or anything you cannot easily recover. Claude Code will do exactly what you tell it. That is its strength and the reason you want to stay deliberate about scope early on.
Structuring Projects So Claude Can Actually Help
How you organize your folders matters more with Claude Code than it does with browser chat. When Claude can read the whole directory, the structure you give it shapes the quality of its output.
A few practices that make Claude more useful across sessions:
- Keep one project per folder — do not drop unrelated files into the same directory. Claude reads context broadly, and mixed files produce mixed output.
- Add a CONTEXT.md or README.md to each project folder — a short file that describes what the project is, who it is for, and what the current status is. Claude will read this automatically and use it to orient its responses.
- Use clear filenames — draft-v1.md, client-brief-acme.txt, sop-onboarding.md. Vague names like notes.txt or file3.docx slow Claude down and produce worse output.
- Version your work — before Claude makes large edits, keep the original. You can ask Claude to save a backup first, or use a simple naming convention like sop-onboarding-original.md.
If you add a CONTEXT.md file, this is the kind of content that makes Claude dramatically more useful from the first message of every session:
# Project: AI Coaching Funnel
Owner: Kashi Keefe
Status: Active build, launching Q3
Goal: Convert cold traffic into booked discovery calls via a quiz funnel
Key files: funnel-copy.md, email-sequence.md, quiz-questions.md
Tone: Direct, warm, no hype, second person
Do not: Add income claims, use the word 'transformative', mention competitorsWhere to Take It From Here
Once you have a working Claude Code session and a clean project structure, the range of what you can delegate expands fast. You can ask Claude to compare two versions of a document and explain which is stronger and why. You can have it draft a full content calendar based on a strategy doc you wrote. You can use it to cross-reference a client intake form against your SOP and flag anything missing.
The pattern that works best: think of Claude as a capable contractor who reads your files, follows your standards, and asks before taking any irreversible action. Your job is to give it clear context, a defined project scope, and a standard it is working against. Its job is to execute without you babysitting every step.
Most people who use Claude Code for a week stop going back to the browser chat for serious work. The difference is not the model. It is the working environment. Get it set up once, build the habit, and the leverage compounds from there.