Every Claude Cowork Feature Explained in Plain English (All 20 of Them)

Claude Cowork has 20 features that most people never fully understand. Skills, connectors, sub agents, MCP servers. Everyone uses these terms like you were born knowing them. You weren't, and that's fine.

So I'm going to walk you through all 20 concepts, starting from the simplest and building up to the stuff that makes Cowork feel less like a chatbot and more like an employee who never asks for PTO. Each one builds on the last.

Prefer to watch? Here's the video version:

1. Workspace Folder

Most people open Cowork and start chatting immediately. That's the wrong move. The first thing you should do is set up your workspace folder. That's the local folder on your computer where every file Claude creates gets stored. Documents, scripts, dashboards, data files, all of it.

Open Cowork, look at the top of the screen, and either pick an existing folder or create a new one. I keep separate folders for every major project (because watching one folder turn into a junk drawer is a personality flaw I've already lived through).

2. CLAUDE.md

There's a single file that controls how Claude behaves in every conversation, and most people don't even know it exists. It's called CLAUDE.md, a markdown file that lives inside your workspace folder. Claude reads it at the start of every session, before you type a word.

You put in who you are, what you do, your rules, your preferences. "Always write in a casual tone." "I run a marketing agency with 12 clients." Once Claude reads that, it stops needing you to repeat yourself. Every conversation starts with full context.

I have different CLAUDE.md files for different projects. My content project knows I run a channel. My business project knows my pricing and services. Set it up once. It changes everything.

3. Global Instructions

CLAUDE.md and global instructions are not the same thing, and confusing them will mess up your entire setup.

CLAUDE.md lives inside one project folder and only applies to that project. Your marketing project has one, your finance project has another. Separate brains. Global instructions apply everywhere: every chat, every project, every session. Think of global instructions as your permanent identity: your name, your communication style, your time zone. Things that never change.

Put the universal stuff in global instructions. Put the project-specific stuff in CLAUDE.md. That's the right setup.

4. Memory

Claude now remembers things you told it weeks ago, and you didn't have to do anything to set it up. Cowork has a built-in memory system. Every conversation, Claude quietly stores the important stuff: your preferences, your workflows, context about your business.

You can see your memory files on the right side of any project. They're just markdown files Claude wrote from your conversations. The best part: you can force it. Say "save that to memory" and Claude locks it in.

My content project remembers my upload schedule, my brand voice, even which topics I've already covered. I never re-explain anything. (The irony of an AI having a better memory than me is not lost on anyone in my household.)

5. Context Window

Ever had Claude forget something you said 20 minutes ago? That's not a bug. That's the context window.

Think of it like a desk. Everything Claude can see right now sits on that desk: your CLAUDE.md, your memory files, every message, every uploaded document, every tool output. Claude currently has a 1 million token context window. Sounds huge. Fills up faster than you think. When it overflows, Claude compresses, pushing out older, less relevant stuff. That's when it forgets.

The fix is simple. Keep conversations focused, start new threads for new tasks, and put the important stuff in CLAUDE.md or memory so it's always loaded first. The critical context never gets pushed off the desk.

6. Multimodal

Everyone keeps saying "multimodal" like you're supposed to know what that means. It's dead simple: Claude can process more than text. Images, screenshots, PDFs, charts, spreadsheets, code files. It sees all of it.

So stop describing the bug. Screenshot the error and drop it in. Stop typing out the table. Upload the CSV. The one thing Claude can't process right now is video. Everything else is fair game.

7. Web Search

Claude's training data has a cutoff date. Anything after it, Claude doesn't know unless it searches. Web search is built right into Cowork. Live pricing, recent news, new tool releases, updated docs. Claude knows when to search automatically, but you can also tell it explicitly: "search for the latest Claude pricing." It pulls real articles, real data, and it cites them.

It's not guessing about things that happened last week. It actually goes and looks.

8. Extended Thinking

There's a way to make Claude think 10x harder about your problem, and most people never turn it on. It's called extended thinking. Give Claude something complex, like a tricky code problem or a multi-step analysis, and it pauses to reason through it step by step before responding.

It takes longer. The output is dramatically better. For simple questions, skip it. For anything where you'd want a human to actually think before answering, turn it on. It's the difference between a rough draft and a final answer.

9. Artifacts

Claude doesn't just give you text. It builds actual working dashboards right inside the chat. Interactive HTML pages, charts, data visualizations, slideshows. You can open them in your browser, save them, share them.

My morning briefing skill creates a full HTML dashboard with my calendar, email highlights, and trending AI news. I've built competitor trackers, content calendars, financial summaries, all as artifacts. And unlike regular Claude chat, artifacts are baked into every Cowork conversation. Ask for something visual and it just builds it.

10. Projects

If you're running everything in one conversation, you're creating chaos. Projects fix that. A project is a separate brain inside Cowork with its own memory, its own instructions, its own conversations, files, and scheduled tasks.

I have one for YouTube content, one for my business, one for my community. They don't bleed into each other. Combine projects with CLAUDE.md files and skills, and each project becomes a fully customized assistant that's an expert in one area of your life.

11. Bash Tool

Claude doesn't just write code. It runs it, right inside Cowork, without you touching a terminal. Whenever Claude needs to process data, resize images, rename files, analyze a spreadsheet, it writes the code and executes it automatically. You watch it happen in real time.

Tell Claude "scan this CSV and flag anything over $10,000" and it writes a Python script, runs it, and hands you the results. No terminal. No installation. No Stack Overflow rabbit holes at 1 a.m.

12. Skills

One command. That's all it takes to trigger an entire workflow. Skills are the reason Cowork is in a different league.

A skill is a markdown file with instructions that tells Claude exactly how to perform a specific workflow. I type "morning briefing" and Claude checks my calendar, pulls my email, scans AI news, writes my priorities, and builds me a dashboard. Same output, same quality, every single time.

Build one once, save it, run it forever. You're programming your own AI employee without writing a line of code.

13. Slash Commands

You can trigger any skill with two keystrokes: slash, then a word. That's it. Slash commands are just trigger words for your skills. /morning-briefing, /invoice-generator, /content-research. Whatever you built.

14. Plugins

People keep confusing plugins and skills. They're not the same thing. A skill is one workflow. A plugin is a bundle of skills packaged together.

Think of a skill as a single app on your phone. A plugin is the download that installs a dozen apps at once. You grab a plugin as a zip file, upload it to Cowork, and every skill inside is immediately available. Anthropic has pre-built plugins for legal, marketing, and design, or you can build your own. Skills do the work. Plugins package it.

15. Connectors

Claude can pull your emails, check your calendar, and post to Slack, all from one chat window. Connectors are how. They're direct integrations between Cowork and your other apps: Gmail, Google Calendar, Slack, QuickBooks, tons more.

My morning briefing pulls from Calendar and Gmail, then sends the summary to Slack. All through connectors. And if the app you need isn't on Anthropic's list, you can add a custom MCP server. Tools like Zapier MCP bridge almost anything. There's really no excuse for Claude not being plugged into your entire workflow.

Connectors are what turn Claude from a chatbot into an operating system for your work.

16. Chrome Extension

Claude can now open websites, click buttons, and fill out forms without you touching your mouse. The Claude Chrome extension is basically a browser agent living inside Chrome. Give it a task and it opens the page, navigates, clicks, reads, and brings the results back to Cowork.

I use it whenever there's no connector for an app. Can't connect directly? Claude goes to the website and does it through the browser. The key setting is "act without asking." Turn it on if you don't want to approve every single click. Just make sure you trust the task first.

17. Computer Use

This one works outside the browser. Claude can access files on your desktop, open applications, drag things between windows. I asked Claude to find a video file on my desktop and drop it into my video editor. It opened the editor, found the file, imported it. I touched nothing.

Honest take: it's slow. Connectors are faster and more reliable, and the Chrome extension handles most browser stuff. But for desktop apps with no API, no connector, and no web version, this is your only option. It's getting better fast.

18. Scheduled Tasks

I wake up every morning to a fully built dashboard from Claude. I didn't ask for it that day. It just runs.

Scheduled tasks replaced my entire automation stack. Take any skill, set a time, and Claude runs it automatically. Every morning at 7 a.m. my briefing fires: calendar, email, AI news, priorities, dashboard, Slack. I'm still in bed. I also run a daily competitor scan, an end-of-day wrap-up, and a community pulse check. All scheduled. All automatic.

This is how you go from using Claude like a chatbot to running it like an employee.

19. Dispatch Mode

You can control Claude on your computer from your phone. At the gym, in the car, anywhere. Dispatch mode turns your phone into a remote control for Cowork on your desktop. Open the Claude app, tap dispatch, type a command. Claude runs it on your computer at home.

Need a file from your desktop uploaded to Google Drive? Send the dispatch. Want your morning briefing early? Send it. If you're away from your desk a lot, this is a massive unlock.

20. Sub Agents

Give Claude one complex task and it splits it across multiple agents working at the same time. Tell it "prep everything for my investor pitch" and agent one does market research, agent two pulls financials, agent three writes the narrative, agent four builds the deck. Simultaneously.

Each agent can use the skills you've already built, and the output comes back as one unified deliverable. This is where everything converges: skills, connectors, artifacts, the bash tool. Sub agents orchestrate all of it.

Now Go Set One Up

That's all 20. If you made it this far, you understand Cowork better than 95% of the people using it.

Pick one concept you haven't tried and set it up today. Start with your CLAUDE.md if you haven't already. That alone changes how every conversation goes. Stop reading. Go build.