What Is an MCP Connector? Permissions Explained for Owners

You may have seen a new button inside an AI chat app that says something like "Add connector" or "Connect your tools." Click it and the app offers to plug into your calendar, your email, your CRM or your file storage. A lot of owners pause right there, and they should. That button decides what an AI tool can see and do inside your business.

This guide explains what an MCP connector is, how the permission step works, and how to make good choices before you click Allow. There is no code here and no vendor pricing. The goal is simple: you should understand what you are approving.

▶ What MCP actually means

MCP stands for Model Context Protocol. Strip away the jargon and it is a shared set of rules that lets an AI model talk to other software in a predictable way. Before standards like this, every AI tool needed its own custom wiring for every app. Now a software company can publish one connector, and any AI app that speaks MCP can use it.

Think of a connector as a key card for an office building. The card itself does nothing until someone programs it. Programming decides which floors it opens, which rooms stay locked, and when the card stops working. The connector is the card. The permissions you approve are the programming.

Once connected, the AI tool can do two kinds of things. It can pull information in, like reading your upcoming appointments or looking up a contact. And, if you allow it, it can push actions out, like creating a task, updating a deal stage or drafting an email. Everything it does flows through what you approved, not around it.

▶ Scopes and permissions, in plain words

When you connect a tool, you usually see a consent screen listing what the AI app is asking for. Those line items are called scopes. A scope is one specific slice of access, such as viewing calendar events or editing contacts. The list is the most important thing on the screen, and most people scroll past it.

Scopes generally fall into a few levels. Reading is the lowest risk: the tool can look but not touch. Writing lets it create or change records. Sending and deleting are the levels to slow down on, because a mistake there reaches your customers or your data directly.

Admin scopes sit at the top. They let a tool change settings, add or remove users, or alter other permissions. That is the highest risk of all, and very few everyday tasks need it. If a connector asks for admin rights, treat that as a question to answer before you approve anything.

Some connectors let you pick scopes one by one. Others bundle them. If a connector asks for more than the job needs, that is a reason to ask why before approving, not after.

• ▶Read: view contacts, calendar events, files or pipeline stages
• ▶Write: create or edit records, add notes, move deals
• ▶Send: email or text people on your behalf
• ▶Delete: remove records, events or files
• ▶Admin: change settings, users or other permissions

▶ You choose which accounts it can touch

Here is the part owners often miss. A connector acts as whoever signed in when it was set up. If you connect using your owner login, the AI tool inherits everything your owner login can do, including billing, user management and every customer record you have ever stored.

That is why the account you sign in with matters as much as the scopes. Many business tools let you create a separate user with limited rights. Connecting through that limited user puts a ceiling on what the AI tool can reach, no matter what it is asked to do in a chat.

Also think about which workspaces and sub-accounts come along. If you run more than one location or brand in the same CRM, decide on purpose which ones the connector can see. Approving access to everything because it was the default is how surprises happen.

• ▶Use a dedicated user for the connector instead of your owner login
• ▶Give that user only the rights the task needs
• ▶Limit it to the specific calendar, pipeline or folder involved
• ▶Keep billing and user management out of reach entirely

▶ Safety habits worth keeping

Connectors are useful, and they are not something to fear. They do deserve the same care you would give a new employee with a key to the office. A few routine habits cover most of the risk.

Start read-only. Ask the AI tool questions about your data for a week or so before allowing it to change anything. You will learn how it interprets requests, and you will spot gaps in your records that would trip up an automation later.

When you do turn on write access, keep a person in the loop for anything customer-facing. Let the tool draft the follow-up email, then read it and press send yourself. Moving a deal stage is low stakes. A message to a customer is not.

• ▶Start with read access and add more only when there is a clear reason
• ▶Review drafts before anything goes to a customer
• ▶Check your connected apps list every month or so and remove what you no longer use
• ▶Remove access the same day an employee who set it up leaves
• ▶Never paste passwords into a chat; use the official sign-in screen
• ▶Keep sensitive files in folders the connector cannot see

▶ Where a connector stops and a running system starts

A connector works while you are in the chat. You ask, it reads or acts, and then it waits for your next request. That is great for looking things up, cleaning records or drafting messages on demand.

It is not the same as a system that runs on its own. Texting back a missed call at night, sending appointment reminders, nudging a quote that went quiet or asking for a review after a finished job all need something that watches for triggers and acts without anyone typing a prompt. Whatever system you use for that work, the same permission thinking applies: give it only the access its job requires.

That is the work Scalogy does. Scalogy is an AI automation firm in Fayetteville, Georgia, that builds and runs these systems on the Scalogy platform, connected to the phone number, calendar, CRM such as HighLevel or HubSpot, and spreadsheets a business already uses.

Most businesses start with one system in the low hundreds per month, and a single automation is usually live within a week. If you want to sort out which tasks belong in a chat connector and which need to run on their own, book a free twenty-minute discovery call.

▶ Frequently asked questions