🧠 How to Actually Use Claude Cowork: Connectors, Skills & Plugins

Most Claude users are doing the same thing.

They open Cowork. They type a prompt. They get an answer. They close the app.

That works. It also leaves 80% of what Claude can actually do completely untouched.

Underneath the chat interface is a three-layer stack — Connectors, Skills, and Plugins — that transforms Claude from a smart assistant into something that runs your actual workflows, connects to your real tools, and executes multi-step work your way. These aren't power-user extras. They're the features that caused $285 billion in software market cap to evaporate the day Anthropic shipped them, because Wall Street understood what most users still haven't: this is when AI stopped being a chatbot and started replacing specialist software.

This guide breaks down all three — what they are, how they interact, and exactly how to set them up.

Before we go section by section, here's the full picture side by side. Screenshot this. Come back to it.

The rest of this guide goes through each row of that table in depth — what it means in practice, how to set it up, and exactly what to type once you do.

Let’s start with connectors.

A connector is a live link between Claude and an external tool. Once it's active, Claude can reach into that tool directly. It doesn't just read — it acts. For instance, it can search your Gmail, draft a reply, pull a file from Drive, post to Slack, and create a task in Linear. You don't paste anything. You don't switch tabs. You just ask.

The difference is hard to overstate. Without a connector, Claude is working blind — it only knows what you tell it. With one, it has eyes inside the tools where your actual work lives.

Anthropic's connector directory covers most of the tools you already use daily:

• Productivity: Google Workspace (Gmail, Drive, Calendar), Microsoft 365 (Outlook, SharePoint, OneDrive), Notion

• Communication: Slack, Teams

• Sales: Apollo, Clay, Outreach

• Legal and Finance: DocuSign, LegalZoom, FactSet

• Other: WordPress, Similarweb, and a growing list

How to set one up (2 minutes)

1. Open Claude Desktop

2. Go to Settings → Connectors

3. Browse the directory

4. Click Add on the connector you want

5. Authenticate with that tool when prompted

That's it — from that point on, Claude has access to whatever you connected — and it uses it automatically when relevant.

One thing to know if you're on a Team or Enterprise plan: some connectors need admin approval before individual users can activate them. Flag it with your IT team first.

Connectors give Claude access to your tools. Skills give Claude access to your brain.

A skill is a plain text file — a Markdown document — where you write down exactly how you want a specific type of work done. Your format. Your standards. Your process. Your tone. Once it's in a skill file and sitting in your Cowork folder, Claude reads it automatically whenever that type of work comes up.

You stop re-explaining yourself every session. Claude just knows.

Why this matters more than prompting

Most people try to solve this problem with longer prompts. They write detailed instructions every single time. They have a folder of saved prompts they paste in. It works, but it's friction and it still produces inconsistent results because the context changes every conversation.

A skill file solves this permanently. You write the instructions once. Claude loads them every time they're relevant. The output stops being generic AI and starts sounding like work you'd actually send.

One well-written skill file is worth more than 50 saved prompts.

What goes inside a skill file

A skill file typically has four things:

1. What the task is
Be specific. Not "write content" but "write newsletter sections in my voice for a professional audience of marketers and founders."

2. Your process
Step by step, how you want Claude to approach the work. If you always start with the problem before the solution, say that. If you never use bullet points in body copy, say that too.

3. What good output looks like
Describe the result you want. Length, tone, structure, and what to avoid. The more concrete, the better.

4. Examples (these are optional but powerful)
Paste in a piece of work you're proud of. Claude will reverse-engineer your style from it.

One thing most people miss: progressive loading

Claude doesn't read every skill file at the start of every session. It reads the title and a short description. When the task you're working on matches a skill, it loads the full file then. This means you can have 20 skill files in your folder without burning through your context window on every conversation.

Three skills every subscriber should build first. These three cover the work most people do most often. Build them in this order:

Skill 1 — Your writing voice

Everything about how you write. Sentence length. Words you never use. Your structure. Your tone. Paste in two or three pieces of writing you're happy with and describe what makes them work. After this skill exists, you never explain your style to Claude again.

What to include:

• Your tone (direct? conversational? formal?)

• Sentence and paragraph length preferences

• Words or phrases you avoid

• 2-3 examples of writing you want Claude to match

Skill 2 — Your document or report format

If you produce recurring documents such as reports, proposals, briefs, client updates — this skill locks in your template. Structure, headings, what goes where, how long each section should be. Claude uses it automatically every time you ask for a document.

What to include:

• Your standard structure and headings

• Formatting preferences

• Tone for this specific document type

• A finished example if you have one

Skill 3 — Your data analysis process

How you want data cleaned, what insights to surface first, how to flag anomalies, what format the output should take. Works with Excel, CSV, any data file you drop into Cowork.

What to include:

• How you want data cleaned and standardised

• What you always look for first (trends, outliers, comparisons)

• How you want findings presented

• Any specific formulas or calculations you use regularly

1. Open any text editor — even a Google Doc works

2. Create a new file and name it SKILL.md

3. Write your instructions using the four sections above

4. Save it and drop it into your Cowork folder

5. Next time you open Cowork, Claude has it

That's the whole process. No installation. No configuration. A text file in a folder.

Pro tip: Start with your writing voice skill first. It has the highest return on investment of anything you can build in Claude. Every piece of content you produce after that benefits from it.

Skills are free to build and free to use. Now for the feature that actually crashed the stock market (with a full breakdown of all 11 official ones and how to build your own from scratch).

A connector gives Claude access to your tools. A skill teaches Claude your process. A plugin bundles both together, adds slash commands and sub-agents, and turns Claude into a specialist you can deploy with one word.

What a plugin actually contains

Every plugin is made up of four components:

1. Skills — the instructions. Markdown files that tell Claude how to do the work your way. This is the same skill file from Section 4, just packaged inside the plugin.

2. Connectors — the tool access. The plugin brings its own connector configuration so you don't have to set it up separately. A sales plugin connects to your CRM automatically. A legal plugin connects to your document management system.

3. Slash commands — one-word triggers for entire workflows. Type /sales:call-prep and Claude kicks off a full pre-call research sequence. Type /marketing:draft-content and it produces a brief, a draft, and three headline variations. One word. Entire workflow.

4. Sub-agents — parallel workers. For complex tasks, Claude splits the work across multiple agents running simultaneously. One researches while another drafts while a third formats. You get the finished output faster without doing any of the coordination.

None of this requires technical knowledge to set up. The whole thing is just files — Markdown and JSON. If you can describe how work should be done, you can build a plugin.

How to install one (in under 60 seconds)

1. Open Claude Desktop and switch to the Cowork tab

2. Click Customize in the left sidebar

3. Click Browse plugins

4. Click Install on whichever plugin you want

5. Type / in the chat to see the slash commands now available

All 11 official Anthropic plugins — what they do and how to use them

Anthropic shipped 11 plugins on January 30, 2026. The same ones their internal teams use. All free. All open source on GitHub. Here's what each one actually does and the prompts worth trying first.

1. Sales

What it does: prospect research, call preparation, outreach drafting, follow-up sequences, and competitive battlecards.

Key slash commands: /sales:call-prep, /sales:prospect-research, /sales:follow-up

Connector it uses: your CRM (Apollo, Clay, Outreach supported)

Try this first:

2. Marketing

What it does: content drafting, SEO audits, email sequences, campaign briefs, competitive analysis, social copy.

Key slash commands: /marketing:draft-content, /seo-audit, /marketing:email-sequence

Try this first:

3. Legal

What it does: contract review, NDA triage, clause flagging, plain-English summaries of legal documents, risk identification.

Key slash commands: /review-contract, /triage-nda, /legal:summarize

This is the plugin that hit Thomson Reuters and LegalZoom hardest. It doesn't replace a lawyer. It does the first pass — flagging unusual clauses, explaining risk in plain English, and surfacing what needs human review.

Try this first:

4. Finance

What it does: financial model building, budget vs actuals analysis, reconciliation, cash flow summaries, scenario modelling.

Key slash commands: /finance:reconcile, /finance:build-model, /finance:variance-analysis

Try this first:

5. Data

What it does: dataset exploration, SQL generation, dashboard building, anomaly detection, chart creation, data cleaning.

Key slash commands: /data:explore, /data:build-dashboard, /data:clean

Try this first:

6. Customer Support

What it does: ticket triage, reply drafting, escalation flagging, FAQ generation, sentiment analysis across support threads.

Key slash commands: /support:triage, /support:draft-reply, /support:weekly-summary

Try this first:

7. Product Management

What it does: feature spec writing, user feedback synthesis, roadmap planning, PRD drafting, and strategy documents.

Key slash commands: /product:draft-prd, /product:synthesize-feedback, /strategy

Try this first:

8. Productivity

What it does: task management, meeting notes to action items, weekly planning, project tracking, inbox zero workflows.

Key slash commands: /productivity:weekly-plan, /productivity:meeting-debrief, /productivity:task-list

Try this first:

9. Enterprise Search

What it does: searches across multiple connected tools simultaneously — Drive, Slack, email, Notion — and synthesizes the results into one answer.

Key slash commands: /search:find, /search:synthesize

Try this first:

10. Biology Research

What it does: literature review, experiment design, research synthesis, scientific writing, protocol documentation.

Key slash commands: /bio:literature-review, /bio:protocol-summary

Built for research teams and life sciences professionals. If that's not your world, skip it.

11. Plugin Create

What it does: builds other plugins. This is the meta-plugin — it walks you through creating a custom plugin from scratch by asking you questions about your workflow.

How to use it: install Plugin Create, then type:

"I want to build a plugin for [describe your workflow]. Ask me questions first to understand what I need before building anything."

Claude generates the full plugin structure — skill files, slash commands, and connector configuration — based on your answers. No code. No files to write manually.

If none of the 11 official plugins match your workflow, build one. It's easier than it sounds.

Install the Plugin ‘Create plugin’ first. Then use this prompt:

Claude asks you:

• What triggers the work

• What steps you follow every time

• What the finished output looks like

• Which tools you need it to connect to

• What you want the slash command to be called

Then it builds the folder structure, writes the skill files, configures the connectors, and creates the slash commands. You review, give feedback, and it adjusts.

The result is a plugin built entirely around your process. Not a generic template you've hacked to fit.

Real example of what's possible: a newsletter workflow plugin with /research-topic that searches the web and pulls from your Drive, /draft-section that writes in your voice using your skill file, and /repurpose that turns the finished piece into a LinkedIn post and three short-form variations. One plugin. Your entire content workflow in three commands.

Here's what Claude doesn't support right now:

• No memory across sessions. Every Cowork session starts fresh. Claude doesn't remember what it did yesterday. Scheduled tasks partially solve this because they run the same task repeatedly, but there's no running memory between separate conversations.

• Desktop only. Cowork runs inside the Claude Desktop app on Mac or Windows. No browser version. No mobile. If your laptop is closed, nothing runs.

• The app has to stay open. Scheduled tasks only run while Claude Desktop is open and your computer is awake. Close the app, and the session dies. This matters if you're building automated workflows and expecting them to run overnight.

• Usage burns faster than regular chat. One complex Cowork session costs significantly more of your usage allocation than a normal conversation. On the Pro plan ($20/month), you'll hit limits if you run heavy workflows daily. If Cowork becomes your main setup, the Max plan ($100/month when paid annually) is worth considering.

• Connector approval on Team and Enterprise plans. Individual users can't always activate connectors themselves. Some require admin approval first. Flag it with your IT team before building workflows that depend on specific connectors.

• Don't use it for regulated workloads. Anthropic is explicit about this. Cowork activity isn't captured in audit logs or compliance exports. If your work has regulatory requirements, this isn't the right tool for that layer yet.

Skip the guide and do this:

1. Set up your first connector today. Go to Settings → Connectors and connect one tool you use every day — Gmail, Drive, or Slack. Two minutes. Immediately changes what Claude can do.

2. Build one skill file this week. Start with your writing voice. Open a text editor, write how you want Claude to write for you, save it as SKILL.md, and drop it in your Cowork folder. Do this once and stop re-explaining yourself forever.

3. Install one plugin. Open Cowork, click Customize, browse the library, and install the one that matches your job. Sales, marketing, legal, finance, data — pick the one closest to your daily work. Type / to see what it can do.

4. Once you have all three running, point them at a workflow you repeat every week. Build a plugin that automates it. The setup takes 45 minutes. The time you get back is permanent.

5. Remember the limits: no memory across sessions, desktop only, app must stay open for scheduled tasks, usage burns faster than regular chat. Build accordingly.

The people who understand this stack now have a compunding advantage over everyone still using Claude as a chatbot. That gap is only going to grow.

You've done the reading. Now go build something.

Until next time,
Kushank @PracticalyAI