How to Connect ChatGPT, Claude or Copilot to Your Business Data: Connectors and MCP in Plain English

Start with the connectors built into a business plan, keep the first connection read-only, and make each person sign in with their own account. ChatGPT, Claude and Microsoft Copilot can all reach your CRM, files and other systems through connectors, many built on the open Model Context Protocol (MCP). The bigger risk is what you let AI touch.

Copying and pasting into a chatbot only gets a business so far. The next step, and the one more owners are asking me about, is letting the AI look things up in your own systems: "Which invoices are past 30 days?" "What did we quote this customer last spring?" "Summarize this week's open jobs."

It isn't just my clients. In an October 13, 2026 Inquirer column, technology consultant Gene Marks wrote that many of his clients are starting to connect Claude or ChatGPT to their data, either with the platforms' out-of-the-box integrations or with MCP. Here's what that means, what each plan allows, and how to do it without handing an AI the keys to everything.

What does connecting AI to your business data actually mean?

It means giving the AI a controlled way to read from, and sometimes act in, the software you already use, instead of only knowing what someone pastes into the chat.

That controlled way is called a connector. A connector sits between the AI and one system, such as your Google Drive, CRM or accounting software. When you ask a question, the AI asks the connector, the connector checks what you're allowed to see, and the answer comes back into the chat.

Many connectors now use the Model Context Protocol (MCP). The official MCP site describes it as an open-source standard for connecting AI applications to external systems, and says assistants such as Claude and ChatGPT support it. You don't need to understand the protocol to use it. What matters for an owner is that one connector can work with more than one AI tool, and that a connector can do more than read: Anthropic's help page says the developer of an MCP server decides what its tools do, which can include creating, modifying or deleting data.

What are the three ways to connect AI to your business systems?

From least to most effort:

1. Use a connector the vendor already built. ChatGPT, Claude and Microsoft all offer ready-made connections to common tools, and many business software companies now publish their own. This is where most small businesses should start. Ask your software vendors what they offer before you pay anyone to build anything.
2. Add a hosted MCP server. If a vendor or a trusted provider runs an MCP server for a tool you use, an admin can add it as a custom connector by pasting its web address and setting up sign-in.
3. Build a custom connector. For a homegrown database or an industry system nobody supports, a developer can build an MCP server for you. Budget for upkeep, not just the build. In the same Inquirer piece, Chop Dawg CEO Joshua Davidson suggests setting aside roughly 10% to 15% a year for upkeep.

Which ChatGPT, Claude and Copilot plans let you connect business data?

Here's what each vendor's own help pages say, as of October 27, 2026. These features are changing quickly, so check the linked page before you buy.

Two details worth knowing. First, both Claude and ChatGPT connect to your server from the vendor's cloud, not from your office. Anthropic says a server on a private network or behind a VPN won't connect unless you allow Anthropic's addresses through your firewall, and OpenAI offers a Secure MCP Tunnel for servers that aren't on the public internet. Second, tool updates behave differently. OpenAI says ChatGPT keeps a frozen snapshot of a server's tools until an admin reviews and publishes an update, while Microsoft says Copilot Studio reflects changes to a server's tools automatically.

Is it safe to connect AI to your business systems?

It can be, if you treat a connector like a new employee's login rather than a convenience. The risks are specific and manageable:

• Shared keys give everyone the same access. Anthropic warns that when a connector uses an API key or other fixed credential on a Team or Enterprise plan, everyone who uses it reaches the service with that same credential. Prefer connectors where each person signs in as themselves.
• Broad permissions widen the damage. The MCP project's security best practices warn against catch-all permissions like "all files" or "admin" and recommend starting with the minimum access a task needs, adding more only when it's required.
• Write actions need a person. A connector that can send, edit or delete can make a mistake at machine speed. Anthropic advises clicking "Allow always" only for servers and tools you trust to run unsupervised, and says its Research mode can call connector tools without asking, so write tools should be turned off there.
• Hidden instructions are real. Anthropic warns that malicious MCP servers may include hidden instructions that try to make Claude perform unintended actions. Connect only to servers from organizations you trust, and review the permissions a connector asks for.
• Local connectors are programs on your computer. The MCP security guidance notes that a local MCP server runs on your machine and, without sandboxing, can run commands with the same privileges as the AI app. Don't install one because someone pasted a setup line into a chat.

None of this is a reason to avoid connecting AI. It's a reason to do it deliberately. If you haven't settled which AI plans your team uses for customer information, read Is It Safe to Put Customer Data in ChatGPT? first; connecting systems builds on that decision.

How should a small business connect its first system?

1. Pick one question worth answering. "What did we invoice last month and what's unpaid?" beats "connect everything." Marks's closing advice fits: choose one process, measure its current time or cost, and check whether the AI improves it.
2. Confirm the plan. Use a business plan your company owns, so an admin controls which connectors are allowed and who can use them.
3. Use the vendor's connector if one exists. Check the AI tool's connector list and your software vendor before considering a custom build.
4. Start read-only, with named sign-ins. Have each person connect with their own account, limit what the connector can see, and leave write actions off for the first few weeks.
5. Write the one-page rule. Which connectors are approved, which systems they can reach, who approved them, and who reviews the list. Pax8's Q3 2026 SMB AI Pulse found 35% of AI-engaged small businesses had discovered AI running without approval in the past six months, so a short written list matters.
6. Add write actions one at a time, with approval. When you're ready for the AI to update a record or draft an email, keep a person approving each action, and keep the vendor's activity logs where your plan offers them.

How does Caprock Agility help connect AI to your business data?

Most of the owners we work with don't need a big integration project. They need to know which system is worth connecting first, what data should stay out of reach, and how to set it up so it still makes sense in six months.

Our AI Readiness Audit ($2,500, 3–5 business days) maps how work actually moves today, including your tools and data, and gives you risk flags alongside ranked AI opportunities, so you know which connection would pay off and which would add risk. If you already know the workflow, EDGE Starter (starting at $8,500, 4–6 weeks) builds one workflow end to end, with training and a Day 30 outcomes review, on accounts your business controls.

You can compare every option on our services and pricing page.