What are MCP connectors?

You already do this the hard way. You open Gmail, copy an email, paste it into the AI, read the reply, then copy that back out. A connector deletes every one of those steps. It lets the AI reach into the tool itself, read what it needs, and act there, while you just ask in plain English. Here is what a connector actually is, and how to turn one on.

What is an MCP connector?

A connector is a link between an AI assistant and a tool you already use, so the AI can read and act inside that tool for you. Turn on the Gmail connector and Claude can read your inbox; turn on the Notion one and it can update your pages. You grant the access once, then you just ask.

The "MCP" part is the plumbing. MCP stands for Model Context Protocol, an open standard for how AI tools plug into your apps. In the app itself you will not see "MCP" anywhere, you will see "Connectors". Most people only ever need the word connector. MCP is just why they all work the same way.

In one line: a connector is a permission you give the AI to use one of your tools, so it can fetch and do things there instead of you copy-pasting.

What can a connector actually do?

A connector lets the AI work inside a tool the way a capable teammate would, using only what you can already see. The clearest way to understand it is by example:

• Gmail or Outlook: "Summarise every email about the Q3 budget from last week." It reads the thread and hands you the summary.
• Google Drive or SharePoint: "Find the documents that mention customer retention and pull out the numbers." It searches your files, not the public web.
• Notion: "Move the onboarding feature to Done and add today's notes." It can read AND update your workspace.
• Slack: "What did the sales channel say about the AcmeCorp deal?" It reads the channel for you.
• GitHub: "Show me the last five pull requests and who reviewed them." It reads your repo.

The pattern is always the same. The connector supplies the AI with live access to your real data, so the answer is about YOUR work, not a generic guess.

How do you add a connector?

You add a connector once, in settings, by signing in the normal way. In Claude, open Settings, go to Connectors, and press the + to add one. You will see a list of built-in connectors (Gmail, Google Drive, Slack, Notion, GitHub and more) plus the option to add a custom one for a company system.

Pick the tool you want and it sends you through the same "Sign in with Google" or "Sign in with Microsoft" screen you have used a hundred times. That sign-in is how you grant access, so you never hand the AI a password. Team and Enterprise accounts work the same way, except an admin usually adds the connector once and each person signs in to their own account.

How do you actually use it in a chat?

This is the step almost everyone misses: you switch the connector on inside the specific chat, every time. Adding it in settings does not turn it on everywhere. In a conversation, press the + near the message box, find Connectors, and toggle on the one you want. Then just ask in plain language.

Turning it on per chat is deliberate. It means the AI only touches your Gmail in the conversation where you asked it to, not in every chat you ever have. If the AI says it cannot see your inbox, this toggle is almost always the reason.

The number one beginner mistake: adding a connector in settings and assuming it is live. It is not, until you toggle it on in that chat.

Is this a Claude-only thing?

No. MCP is an open standard, not a Claude feature, so the same idea works across the major AI tools. Anthropic created it, then handed it to the Linux Foundation in December 2025, and OpenAI, Google and Microsoft have all adopted it. ChatGPT, Gemini and Copilot support connectors built on the same protocol.

For you that means one thing worth knowing: connectors are not a niche power-user trick anymore, they are becoming the standard way every AI assistant plugs into your tools. Learning them once carries across whichever AI you end up using.

Is it safe to connect my accounts?

Yes, within limits you control, because a connector can only ever do what YOU can already do. Three things make it safe:

1. It signs in, it never sees your password. The connect step uses the same secure sign-in (OAuth) as any "log in with Google" button, and you grant specific access, like "read Gmail", not full control.
2. It inherits your permissions. If you cannot open a colleague's private file, neither can the AI through your connector. It is you, extended, not a new person with new access.
3. You can switch it off any time. Toggle it off in a chat, or disconnect it entirely in settings, and the access is gone.

The sensible habit is to read what a connector asks for when you add it, the same way you would glance at what a new app wants before you tap "Allow".

Before you switch one on

• Do you actually use the tool enough for live access to save you time? Start with the one you copy-paste from most.
• Are you comfortable with what it is asking to access? If yes, connect it. If not, skip it.

A connector turns the AI from something you paste into, into something that works where your work already lives. Add it once in settings, remember to toggle it on in the chat, and start with the tool you touch every day. The starter kit below gives you the five to switch on first, and the exact first thing to ask each one.