A client told me last month that they were "already using Claude." Great. I asked how.
They opened their laptop and showed me. They were copying text from their CRM, pasting it into Claude, asking a question, then copying the answer back out. Every single time.
That's Claude on its own. It works. But there are two other versions of "using Claude" that would eliminate most of that manual work. Most people don't know those versions exist, or they've heard of them but aren't sure what's actually different.
Think of it like a car with different trims. Same make and model on the outside. Completely different experience depending on what's under the hood.
Version 1: Claude on Its Own (Base Trim)
This is the one most people start with. You go to claude.ai, you type, Claude answers. You paste in a document, Claude reads it. You describe a problem, Claude helps.
Claude here is smart but isolated. It can only work with what you put in front of it. It has no idea what's in your email, your CRM, your spreadsheets, or your files unless you copy that content in yourself. Every session starts fresh. Nothing carries over.
Think of it like calling a really knowledgeable consultant who just walked in off the street. They're brilliant. They can help you with almost anything. But they don't know your business, your clients, or your history. You have to brief them every single time before they can do anything useful.
This is still genuinely valuable for drafting, thinking, research, writing, summarizing, and analyzing things you share with it. For a lot of tasks, this is all you need.
The limitation is the manual handoff. Every time you want Claude to work with real information from your business, someone has to go get that information and hand it over.
Version 2: Claude With Connected Apps (Mid Trim)
Claude.ai lets you connect certain apps directly in your settings. Google Drive, Google Docs, Outlook, and others depending on what's available. Once connected, Claude can look things up in those places without you having to copy anything over.
Same brilliant consultant, but now you've given them a key to the filing room. They can go pull what they need when you ask a question. You still direct everything. You're still in the conversation. But they're not working blind anymore.
You can ask "what did we discuss with this client last month" and Claude can actually look through your connected notes or emails to find it. You can say "draft a reply based on the email I got from John this morning" and if your email is connected, Claude reads it first. The manual copy-paste step is gone for the tools that are linked.
The catch is that the list of supported connections is limited to what Claude.ai has pre-built. If your CRM isn't on the list, or your project management tool isn't supported, you're still doing it manually for those.
And even with connected apps, Claude isn't doing anything on its own. It's not monitoring your inbox, taking initiative, or running processes in the background. It waits for you. It just shows up better prepared.
Version 3: Claude With MCP (Full Build)
MCP (Model Context Protocol) is how you go beyond the pre-built connection list. It's a standard that lets developers build custom connections between Claude and basically any tool or system.
If Version 2 is giving the consultant a key to the filing room, Version 3 is giving them a full office badge, system login, and access to every room in the building. Including the rooms that weren't on the original list.
Your specific CRM. Your internal database. Your custom apps. Your industry software. Your scheduling system. If it has an API (a way for software to talk to other software), MCP can bridge it to Claude.
This is what people usually mean when they say they want AI "connected to their business systems." Not the short list of pre-approved apps. Their actual setup.
The tradeoff is that it takes more to build. Someone has to configure the MCP server (the piece that does the translating between Claude and your tool). It's not a settings toggle. It's a real project.
MCP also opens the door to agents. When Claude has custom tool access via MCP and the ability to take action (not just answer questions), you can build something that works without you directing every step. That's a separate topic. But MCP is what makes it possible.
The Part That Trips People Up
All three of these look similar from the outside. Same chat interface. Same name. But they're doing very different things.
It's like the difference between a flip phone, a smartphone, and a smartphone with every app, integration, and automation set up. They're all phones. You can make calls on all of them. But the gap in what they actually do for your day is enormous.
Version 1: Claude knows what you told it this session.
Version 2: Claude can look things up in approved apps when you ask.
Version 3: Claude can connect to your actual systems and (if built that way) take action in them.
People sometimes ask for MCP when what they actually want is Version 2. They just want to stop copy-pasting from Google Drive. That's a settings change, not a development project.
And people sometimes think they need Version 3 when they'd get 80% of the value just from using Version 1 more intentionally with better prompts. Sometimes the friction isn't the tool. It's the habit around it.
Which One Is Right for You Right Now?
**Version 1** makes sense if your use cases are mostly writing, thinking, summarizing, or analyzing content you can easily share. No setup needed. Value shows up the same day you start using it consistently.
**Version 2** makes sense if you're already using Claude regularly and the copy-paste step is the main friction. Check what's supported in your settings, connect what fits, and see if it changes things. Still no developer needed.
**Version 3** makes sense if your most important systems aren't on the pre-built list, if you need Claude working with data that lives somewhere specific to your business, or if you're building toward something more automated. This is a real project. It needs time and budget.
Most small businesses right now are somewhere between Version 1 and Version 2. That's not a problem. That's where the clearest ROI is and where the risk is lowest. Version 3 is coming for more businesses in the next year or two. It doesn't have to be your first move, and honestly for most people it shouldn't be.
A Note on the Naming
The terminology in this space is genuinely confusing because different tools use similar words to mean different things. "Connected" in one platform means something completely different than "connected" in another. MCP is a specific technical standard that not every AI tool supports yet.
When someone tells me they want AI "connected to their systems," the first thing I do is find out what they actually mean by connected. Do they want to read from their systems? Write to them? Take action automatically? That one word covers an enormous range.
Usually once we talk through what the actual goal is, the right version becomes obvious pretty fast.
Michelle Onizuka is co-founder and Systems Architect at Onizuka Studio. She helps small and mid-size businesses figure out which AI setup actually fits their operation before spending money building the wrong one.
[Start with an Automation Audit](/automation-audit/) if you want a straight answer on where to begin.