Getting Started with Claude Cowork
A Simple Guide for getting started with Claude cowork
Part 1: What Is Cowork (And How Is It Different)?
You already know Claude — the AI you chat with to get answers, write things, think through problems.
Cowork is different. Instead of a back-and-forth conversation, you give Claude a task, point it at a folder on your computer, and let it do the work — reading files, creating files, organizing things — while you go do something else.
| Regular Claude Chat | Claude Cowork |
|---|---|
| You type, Claude responds | You describe an outcome, Claude executes |
| You apply changes yourself | Claude writes changes directly to your files |
| One prompt, one response | Multi-step tasks, runs until done |
| Like texting a smart friend | Like handing off to a capable coworker |
The key mindset shift: You’re not prompting for answers. You’re delegating tasks.
Part 2: Setup (Do This First)
Step 1 - Install Claude Desktop
Cowork only works in the desktop app (not the web browser).
Download at: claude.com/download → Install for Mac or Windows.
Step 2 - Create Your Cowork Folder
Create a dedicated folder on your computer. This is the only place Claude can see and touch files.
📁 cowork/
📁 inputs/ ← drop files here for Claude to work on
📁 outputs/ ← Claude puts finished work here
📁 tasks/ ← your MD task files live here
⚠️ Don’t point Cowork at your entire Desktop or Documents folder. Keep it scoped. You control what Claude can access.
Step 3 - Open Cowork in the App
- Open Claude Desktop
- Click the “Cowork” tab at the top
- Click “Add Folder” → select your
cowork/folder - You’re ready.
Part 3: Your First Task
Before anything fancy, just talk to it. Plain English.
Try this:
Look at the files in my inputs/ folder.
Create a summary of each file and save the results
to outputs/file-summaries.md
That’s it. Watch it work. Come back when it’s done.
Rules for good task descriptions:
- Be specific about where to read from and where to write to
- Tell Claude the output format you want (.md, .xlsx, .docx, etc.)
- One clear goal per task
Part 4: MD Files - Your Reusable Task Templates
This is where Cowork gets powerful.
An MD file (markdown file, .md) is a plain text file with instructions. Write what you want Claude to do once, save it, reuse it forever. Think of it as a standing operating procedure for Claude.
Why use MD files instead of typing tasks each time?
- Write the instruction once, perfectly — no need to re-explain
- Share the same file with teammates so everyone runs tasks the same way
- Claude reads it as a briefing doc before starting work
What an MD File Looks Like
# Task: Weekly File Organizer
## What This Does
Organizes new files dropped into inputs/ and sorts them into labeled subfolders.
## Instructions
1. Scan all files in the `inputs/` folder
2. Sort them into subfolders by type:
- `inputs/documents/` → PDF, DOCX, TXT, MD
- `inputs/images/` → JPG, PNG, GIF
- `inputs/data/` → CSV, XLSX
- `inputs/other/` → everything else
3. Rename each file: `YYYY-MM-DD_original-name.ext`
4. Create `outputs/organization-log.md` listing:
- Total files processed
- Where each file was moved
- Any files you weren't sure about (flag them)
## Do Not
- Delete any files
- Overwrite existing files in subfolders
## Output
Save the log to `outputs/organization-log.md`
Save this as tasks/file-organizer.md.
To run it:
Read tasks/file-organizer.md and execute the instructions.
Part 5: The MD File Formula
Every good task MD file follows this structure:
# Task: [Name of the task]
## What This Does
One sentence. What is the purpose of this task?
## Inputs
What files or data does Claude need?
Where are they located?
## Instructions
Step-by-step. Be specific. Claude follows these literally.
## Rules / Do Not
What should Claude avoid? What are the guardrails?
## Output
What file(s) should Claude create? Where? What format?
The more specific you are, the better the output. Vague instructions = vague results.
Part 6: Practice Tasks
Each of you has one task. Create the MD file, save it in tasks/, run it.
🟣 Product Notes Organizer
Scenario: You have a folder of messy product notes, random docs, and meeting summaries. You want a clean, structured brief.
Create tasks/product-brief-generator.md that tells Claude to:
- Read all
.txtand.mdfiles ininputs/product-notes/ - Extract: key themes, open questions, decisions made, action items
- Output a clean brief at
outputs/product-brief.md - Use headers: Summary / Key Decisions / Open Questions / Next Steps
Run it:
Read tasks/product-brief-generator.md and execute the instructions.
🟡 Data Cleanup Pipeline
Scenario: You have CSV exports from different sources, inconsistently formatted. You want them merged and cleaned.
Create tasks/csv-cleanup.md that tells Claude to:
- Read all
.csvfiles ininputs/data/ - Standardize date columns to
YYYY-MM-DD - Remove duplicate rows
- Merge into one file:
outputs/merged-data.csv - Log what it did at
outputs/cleanup-report.md(files processed, rows removed, issues flagged)
Run it:
Read tasks/csv-cleanup.md and execute the instructions.
🔵 Meeting Notes → Action Items
Scenario: Folder of meeting notes saved as text files. You want action items extracted and tracked.
Create tasks/action-item-extractor.md that tells Claude to:
- Read all files in
inputs/meeting-notes/ - For each file, extract: action items, owner (if mentioned), due date (if mentioned)
- Compile into
outputs/action-tracker.md - Format as a table: Action Item | Owner | Due Date | Source File
- Flag action items where the owner is unclear
Run it:
Read tasks/action-item-extractor.md and execute the instructions.
Part 7: Scheduling Tasks
Once your MD files are working, you can automate them to run on a schedule.
In any Cowork task, type:
/schedule
Set it to run daily, weekly, or on demand. Claude will execute your MD file automatically.
Scheduled tasks only run while your computer is awake and Claude Desktop is open.
Part 8: Ground Rules
Before handing anything to Cowork:
- Start scoped. One folder. Not your whole computer.
- Back up first. Cowork can move and edit files. For anything important, have a backup.
- Cowork ≠ chat. Don’t use it for a quick question. Use it when you need files produced.
- Review outputs. Claude is fast, not infallible. Spot-check the results.
- No sensitive data. Passwords, financial data, confidential info - keep it out of the Cowork folder.
Quick Reference Card
Start a task from an MD file:
Read tasks/[filename].md and execute the instructions.
MD file template:
# Task: [Name]
## What This Does
## Inputs
## Instructions
## Rules / Do Not
## Output
Folder structure:
📁 cowork/
📁 inputs/
📁 outputs/
📁 tasks/ ← your .md files live here
The best way to learn Cowork is to break something small. Start with a test folder, run a low-stakes task, and see what comes back. The first time it works, you’ll immediately think of five more things to hand off.