What is MCP, in plain English?
MCP is a common way for AI assistants to plug into other software. Before it existed, every connection between an assistant and a business system had to be built separately. MCP gives both sides a shared set of rules, so one connector can work with any assistant that supports the standard. The glossary entry for MCP has the short version.
There are three parts to know:
- The assistant (the client) is the app you type into, such as Claude or ChatGPT.
- The MCP server is a small piece of software that sits in front of a system, such as your CRM or your accounting records, and offers the assistant a list of things it can do.
- Tools are the individual actions on that list, such as “search customers”, “show overdue invoices” or “create a task”.
When you ask a question, the assistant decides which tools to call, the server runs them and hands back the results, and the assistant writes its answer from what came back.
Why does MCP matter for business data?
MCP matters because it lets an assistant work with your actual records instead of guessing. Without a connection, an assistant can only work from what you paste into the chat. With one, you can ask “which deals did we win last month that haven’t been invoiced?” and get an answer built from your own data.
That is also why it needs care. The assistant is now reaching into systems that hold customer details, prices, invoices and staff information. The questions that matter are the same ones you would ask before giving a new employee a login: what can they see, what can they change, and who checks their work.
Bear in mind that whatever a tool returns is passed to the AI model to produce the reply. The provider’s terms on data handling apply to it, so read them.
What is the difference between read tools and write tools?
Read tools look things up. Write tools change something. This is the most useful distinction to make when you decide what to connect.
| Read tools | Write tools | |
|---|---|---|
| Examples | Search leads, list overdue invoices, summarise a customer | Add a note, create a task, draft a letter, update a record |
| Worst case if misused | Someone sees data they shouldn’t | A record is changed or created wrongly |
| What to check | That results respect the user’s permissions | Which actions are allowed, what is logged, and whether it can be undone |
| Sensible starting point | Turn on first, for people who already have access | Turn on selectively, once you trust the read side |
Some actions should not be available through an assistant at all, or only in a limited form. Sending an email to a customer, posting a letter or raising an invoice are examples where a firm may reasonably insist that a person clicks the button inside the system itself.
Whose permissions does the assistant use?
The safest answer is that the assistant acts as you, with your permissions and nothing more. This is the single most important question to ask about any MCP connection.
Some connections use one shared account with broad access, often set up by whoever installed it. That means a junior member of staff can ask the assistant for data they could never open themselves, and the assistant will answer. A better design has each person connect with their own login, so a sales executive’s assistant sees what that sales executive sees, and a finance manager’s sees what the finance manager sees.
What should you ask before connecting an assistant?
Ask these questions of whoever provides the connector, whether that is a software vendor or your own developer:
- Does each person connect as themselves? Or does everyone share one account?
- Which tools read and which tools write? You should get a plain list.
- Are any actions blocked or limited through the assistant? Which ones, and why?
- Is every write recorded? You should be able to see who changed what, and when.
- Can access be revoked? Both by the user and by an administrator, straight away.
- Where does the data go? Which AI provider receives tool results, and under which terms?
- Can you start with read-only access? A good provider will let you.
If the answers are vague, treat the connection as having full access to everything the underlying account can reach.
A worked example: two ways to connect
This example is illustrative. Take a 25-person agency where the founder wants the team to ask Claude about the sales pipeline.
Option one is a single connector set up under the founder’s admin account. It is quick. But now a new account manager can ask “what did we invoice each client last year?” and get the full figures, including clients they have never worked on.
Option two has each person connect with their own login. The account manager asks the same question and sees only their own clients, because that is all their login can see. The founder asks it and sees everything. Nobody had to write a separate rule for the assistant, because it inherits the rules that already exist.
The second takes a few more minutes per person to set up. It is the one to insist on.
Do you need a developer to use MCP?
You do not need a developer to use an MCP connection that someone else has built. In most assistants, adding one is a settings step: you add the connector, sign in, and approve what it can access. You would need a developer to build an MCP server for a system that does not have one yet.
Where OpsUPLOOP fits, and where it doesn’t
OpsUPLOOP works with Claude and ChatGPT through MCP. Each person connects with their own account, and the assistant gets only what that person is allowed to see and do in OpsUPLOOP. Some actions are restricted or unavailable through an assistant: for example, a billable created through an assistant is never raised as an invoice automatically, and letters are reviewed by a person before anything is posted. Each customer gets their own instance, configured to their firm. The OpsUPLOOP platform page lists what you can ask.
OpsUPLOOP is not a general MCP gateway for your other systems, and a connection does not make an assistant’s answers correct. The assistant can still misread a question or summarise badly, so decisions that affect customers or money stay with a person. If you are looking for AI to monitor servers or IT incidents, that is a different category; see AIOps vs AI for business operations.
Other questions
Does MCP work with ChatGPT as well as Claude?
Yes. MCP was introduced by Anthropic, the company behind Claude, and has since been adopted by other providers, including OpenAI. Support differs between plans and apps, so check what your own subscription allows.
Is an MCP server the same as an API?
Not quite. An API is how one piece of software talks to another. An MCP server usually sits on top of an existing API and describes its actions in a way an AI assistant can discover and use.
Can I switch an MCP connection off?
You should be able to. A well-built connection lets you disconnect the assistant or revoke its access at any time, and an administrator should be able to do the same for any user.
Does connecting an assistant train the AI on my data?
That depends on your AI provider and your plan, not on MCP. Read your provider's data-use terms, and check whether business plans handle your data differently from personal ones.