This is a live walkthrough of Claude Cowork doing real work — not a demo task like reorganizing a spreadsheet, but actual business output: importing and customizing vendor documentation into a branded knowledge base, completely autonomously. This tutorial explains the interface, the difference between skills and MCP connectors, and how the human-in-the-loop model works in practice.

The Claude Cowork Interface

Claude Cowork looks similar to Claude chat, with a few key differences:

• Left panel: Your conversation/task history
• Bottom: The chat/task input — instead of “New Chat” it says “New Task”
• Right panel: Progress, working context, and Suggested Connectors — this is where you see which tools (MCPs) the agent is using in real time

Claude Cowork (and Claude Skills) is now available on the Pro plan ($20/month) — not just the Max plan.

Skills vs. MCP Connectors — The Key Distinction

These two things work together, and understanding the difference is what makes the whole system click:

Skills = Employees. A skill is a trained AI agent with a specific job: a document importer, a script writer, a community post creator. The skill knows what to do, how to do it your way, and what the output should look like. You create them once and use them whenever you need that job done.

MCP Connectors = Tools. An employee without tools can only think — they can’t act. MCP connectors give your skills the ability to actually do things: read a website, publish to WordPress, send an email campaign, post to social media, manage a community space.

The combination — skill + connector — is what produces autonomous output. The skill knows the job. The tool does the work.

Types of MCP Connectors Available

Claude Cowork gives you several layers of connectivity:

• Default connectors: Google Drive, email, calendar, GitHub — available out of the box
• Desktop extensions: Custom MCPs that run locally on your computer — including ones you build yourself (like the FluentCRM connector James built)
• Browse extensions: A growing library of publicly available connectors — Asana, Canva, Notion, HubSpot, Figma, ClickUp, Intercom, Stripe, PayPal, Vercel, Udemy Business, and hundreds more being added continuously

For the demo in this video, two connectors are active: the web browser (to read the FluentCRM documentation site) and AI Engine (to publish directly to WordPress).

The Live Demo: Doc Translator + AI Engine

The task: take FluentCRM’s official documentation (120 articles), identify the ones most relevant to education businesses, customize them for campus context, and import them as draft BetterDocs articles on the TrainingSites WordPress site.

Step 1 — Ask the agent what’s there. James asks Cowork to review the FluentCRM docs and report back. The agent browses the site as if it were a person, reads the structure, and returns a categorized summary: Getting Started, Automations, Contacts, Emails, Forms, Integrations — 120 articles total.

Step 2 — Narrow the scope. James asks which categories are most useful for education businesses. The agent identifies Automations, Contacts, Emails, and Forms as the top priorities — skipping technical integrations that aren’t relevant to a campus operator.

Step 3 — Give the instruction. “Import categories 1 through 4.” The agent creates a to-do list in the right panel and starts working. James steps away.

Step 4 — Human-in-the-loop check-in. The agent completes one article and returns with a progress report and three options: (1) continue manually, one-by-one for high quality — estimated 4-6 hours for 21 articles; (2) batch import basic titles and structure; (3) identify the 5-7 most important articles and do those with full quality. James picks option 3.

Step 5 — Review the output. The first completed draft — “Introduction to Student Journey Workflows” — is in WordPress as a BetterDocs draft with the correct category, appropriate tags, and campus-specific language throughout. Generic FluentCRM terms like “subscriber” have been translated to “member,” “campus,” “student journey.”

What This Replaces

To do this manually: read each vendor doc, rewrite it for your context, copy-paste into WordPress, set the category, add the tags, save as draft. For 21 articles, that’s half a day of focused work — or you hire a VA and spend time briefing and reviewing.

With Cowork: describe the job, set the constraints, check in at decision points. The agent does the reading, writing, categorizing, tagging, and publishing. You review the output.

The Bigger Picture

This is not about prompting. It’s about understanding the workflows and processes your business needs, making sure you have the right skills to handle them, and connecting the tools those skills need to complete the work.

The shift: from “how do I do this?” to “what do I want done — and which employee handles it?”

Join the community and access the full skills library at trainingsites.io/join.