Three Products, One Name, Zero Syncing
This is part fourteen of our Putting AI to Work in Your Tour Business series. Watch the full series here and read part one here.
Anthropic calls its AI Claude. It also calls its terminal tool for developers Claude Code. And it calls its desktop app Claude Cowork. Same word up front, three separate products, and none of them share anything with each other.
That is not a minor detail. It is the thing that trips people up first, and it is worth getting straight before you touch a single setting.
Claude.ai is the web chat. Open a browser, have a conversation. If you have used ChatGPT or Gemini, you already know how this works. It syncs across the phone app and the browser.
Claude Code runs in a terminal window. It is built for people writing software. If that is not you, you can safely ignore it for now. Kelsey mentions it only so the name does not confuse you later.
Claude Cowork is the desktop app. And here is the useful part: Cowork and Claude Code are more or less the same tool underneath. Cowork is a friendlier wrapper around the same capability.
“Claude Cowork is really Claude Code, but for the rest of us.”
Now, the piece that catches people. Your Cowork sessions and tasks do not sync to Claude chat. Your Claude chat projects do not appear in Cowork. A connector plugged into Claude.ai is not automatically plugged into Cowork. Three distinct products, three sets of settings.
It takes some adjustment. The trade-off is a lot more capability.
Privacy First, Before Anything Else
Open the menu at the bottom left, click your name, and you are in account settings. Head to privacy.
Two toggles matter here. Turn off location metadata. Turn off help improve Claude. That second one keeps your chats and your work from being used to train future models. If you are about to load a folder with your pricing, your client list, and your operating context into this thing, that is a five-second decision with a long shelf life.
Usage: The Setting You Will Care About More Than You Expect
Billing lets you move between plans. Usage is where things get practical.
Cowork runs on a five-hour window with a usage allowance, plus a weekly limit. Both are measured in tokens. Once you start building an AI brain, creating tasks, and spinning up agents that take action for you, those limits stop being abstract.
A practical pattern: use Opus to build the plan or tackle something complex, then switch to Sonnet to execute. Sonnet is the more efficient model, so you can run it far longer before you hit a ceiling.
“Use Opus to create the plan. Use Sonnet to do the execution.”
There is also an extra usage toggle that keeps Claude running past a limit. Useful when something is genuinely high priority. Also a fast way to spend more than you meant to. You can switch it off entirely, or set a monthly spend cap.
If you are in the brain-building phase, watch this screen. Some operators upgrade a plan temporarily while they build out their brain and connectors, then drop back down afterward.
Capabilities and Connectors
Under capabilities, turn on search and reference chats, turn on generate memory from chat, and give it most of what is offered below.
Then connectors. These are the nervous system of your AI brain.
“Connectors are like the nervous system to your AI brain.”
A connector gives Cowork the ability to reach into a piece of software, read information, and in many cases change things and take action for you. Each one is managed individually, so you can start conservative and require permission before anything happens. That is where we recommend starting.
As you get more comfortable, you give it more room. Kelsey now grants a fair amount of independent action, and still asks Cowork to check in or break a project into steps when it matters. Control does not disappear because capability grew.
Hit the plus icon and browse connectors. Viator and TripAdvisor are already there. Scroll down and you will find the desktop extensions, which are permissions for tools sitting on your actual computer. iMessage, Apple Notes, Google Chrome. The PDF extension is worth adding for anyone who handles a lot of documents.
Build Your AI Brain: Context Engineering for Tour Operators
What an MCP Server Actually Is
Picture two sides. On one side, every piece of software in the world, each speaking its own language through its own API. On the other side, the AI models. They cannot talk to each other directly.
An MCP server sits in the middle as a translator. It lets software on one side communicate with an AI model on the other, so Claude can pull information from your tools, work with it, and act inside them.
The Zapier MCP server is one worth knowing. Zapier has already spent years learning to speak the language of thousands of applications. Connect a Zapier account, spin up an MCP server, and Claude can reach anything you have wired into Zapier.
There is a cost. Zapier acts as a middleman, and that means additional token usage. Which is why more companies are now building their own MCP servers so models can connect directly. Expect booking software in the tour and activity sector to follow. In the meantime, if you are stuck connecting a piece of technology, Zapier MCP is your best friend.
“If you don’t empower the AI to talk to your tools, you’ll be pretty limited in what you can accomplish.”
Skills, Plugins, and the Difference Between Them
A skill is a workflow. Specific instructions, specific tools, clear rules on what to do and what not to do, so you get consistent output every time you run it.
Anthropic and its partners publish skills you can add from the browse skills menu. Install the skill creator first. It exists because Anthropic knew people would want to build their own workflows, and it makes every skill you build after it better.
Plugins are the next size up. The data plugin, the productivity plugin, and for business owners, possibly the legal plugin.
Here is the analogy that made it click. A skill is an SOP. Narrow scope, specific outcome, defined process. A plugin is a bundle of skills, closer to a role. Eventually these bundles may become full agents. Spin up a sales role. Spin up an operations role. Underneath, it is still a stack of skills.
“A skill is an SOP. A plugin is almost like a role.”
You can also download skills from the growing number of skill databases online. Be a little careful here. A skill is ultimately a text file with instructions, and it is going into your brain. Kelsey stays conservative about dropping downloaded skills straight in. You can always ask your AI to analyze a skill before you install it.
The One Folder Rule
Scroll to the desktop app settings and you will find toggles for running Cowork at startup, keeping your computer awake so it can work while you sleep, and letting it access your browser. Most of these start switched off. Come back to them when you are ready.
Then go to new task, and pay attention to the folder.
Always use the same folder. Every single time. Use the dropdown arrow and hit the star to make it your default.
You will hear other Cowork tutorials suggest different folders for different work. We recommend against it, and the reason is simple. The AI Brain Builder prompt creates a full folder structure that holds your memory, your instructions, and the deep context about your business. Point Cowork at a different folder and none of that comes with you. No instructions. No context. A powerful tool with amnesia.
“If we’re using different folders, you’re not going to have that context.”
Same logic applies to Cowork projects. Each project spins up its own separate folder, which is exactly what we are trying to avoid. Yes, you can still run projects. They just all live in the one central folder.
Build Your AI Brain: Context Engineering for Tour Operators
Scheduled Tasks and What to Watch While It Works
Scheduled tasks live in the menu, though you can also just ask inside a task to schedule something. The menu is useful because it shows what is running and what is active.
Click new task and you can set frequency: manual, hourly, daily, weekdays, weekly. Point it at the same folder. Give it permission to change files inside that folder, otherwise it cannot do the work. Choose your model, bottom right. Sonnet will handle most scheduled work fine.
Once a task is running, the right-hand panel is where the real value sits. The progress section breaks complex work into steps, and you can watch multiple agents working at once. Below that is your sync folder, showing every file it referenced, read, or created.
That panel is how you verify. Did it actually open the business context files? Did it use the connector you asked it to use? Expand any section with a chevron and you see the work. Kelsey uses this constantly to confirm the AI actually knows what it is talking about before trusting the output.
At the bottom, context shows which skill and which tools were used. Outputs get quick links so you can open a file in Finder or a browser.
A Workstation, Not a Lock-in
If you have followed this series and built your brain, Cowork becomes your workstation. It is where the capability we cover in the videos ahead actually gets used.
You do not need to be technical. Ask Cowork how to use Cowork. Kelsey does this daily. It breaks things into bite-sized pieces without making you feel stupid.
And the brain itself stays portable. It is markdown and text files. If a better tool arrives next year, the brain moves with you.
“Once we’ve created the AI brain, it’s just text files. We can move it to another tool whenever a better one shows up.”
If you are nervous about setting this up, or you got stuck partway through, book a free strategy call with one of our coaches at https://guestfocus.com/checkin. We will get you unstuck and show you what members are doing with it.



