If you’ve ever wished your AI assistant could actually check your Slack, look at your Google Drive, or update a ticket in Jira — without you copying and pasting anything — that wish already has a name. It’s called MCP, and inside Claude it shows up as something called “Connectors.”
This post skips the jargon and explains what these two things actually are, how they’re different from each other, when you’d want to use them, and — just as importantly — what changes in your day-to-day work once you flip that switch on.
Let’s start with a simple picture
Imagine Claude as a very smart, very fast assistant sitting at a desk. By default, that assistant only knows what you tell it in the conversation. It has no idea what’s in your inbox, your calendar, your project tracker, or your files sitting in Google Drive. If you want it to help with something in one of those places, you have to open that app yourself, copy the relevant bit, and paste it into the chat.
Now imagine giving that assistant a set of keys — one key per app. Suddenly it can walk over to your Slack, your Gmail, your Drive, or your company’s internal tool, look at what’s there, and even take action, like sending a message or creating a task. That’s the entire idea. The “keys” are called Connectors, and the standard that makes the keys fit the locks is called MCP.
What is MCP, in plain words?
MCP stands for Model Context Protocol. Don’t worry about the name. Think of it as a universal plug shape. It’s an open standard Anthropic introduced in November 2024, and instead of every app needing its own one-off, custom-built integration, MCP gives any service a single, common way to make itself available to an AI system like Claude.
It’s the same idea as a USB port. Before USB, every device — printer, mouse, camera — needed its own special cable and its own special socket. USB gave everyone one shape to build around. MCP is doing that for AI: instead of Anthropic having to hand-build a custom bridge to every single app in the world, any company can build one MCP-shaped bridge to their own product, and it’ll work with Claude.
MCP is not owned by one single company forever, and it isn’t only for Claude. Because it’s a shared standard, a connector built on MCP can work across more than one AI assistant, not just Claude. It’s genuinely becoming a common language for “AI talks to your apps.”
So what is a “Connector,” then?
A Connector is simply the actual bridge that’s built using MCP, sitting inside your Claude account. If MCP is the plug shape, a Connector is the specific cable plugged into a specific device — Slack, Google Drive, Jira, Notion, GitHub, Figma, and dozens of others.
Once you turn a connector on in a conversation, Claude will reach for it by itself whenever your request calls for it — you don’t have to say “check Jira” every time. Ask something like “what’s blocking the release?” and if Jira is connected, Claude already knows to go look.
These connectors work the same way whether you’re on the claude.ai website, the Claude desktop app, Claude Code, Cowork, or the API — so it isn’t a feature tucked away in one corner, it follows you across however you use Claude.
Two flavours of connector
Directory connectors
These are the ready-made, verified integrations Anthropic maintains a Connectors Directory for — Slack, Google Drive, Jira, and similar well-known tools. You just find them and click connect.
Custom connectors
These let you link Claude to any external service running its own remote MCP server — including internal company tools, niche apps not in the directory, or something your own team built.
MCP is the standard. Connectors are what you actually click.
The real difference, side by side
| MCP | Connector |
|---|---|
| The rulebook / protocol that defines how an AI and an app can talk to each other | An actual, working link between Claude and one specific app, built using that rulebook |
| You never “click” on MCP itself | You click “Connect” on a connector, sign in, and it’s live |
| Mostly relevant to developers building integrations | Relevant to everyday users — this is the thing you actually use |
| One protocol, works across different AI assistants | One connector, tied to one specific app (e.g. just Slack, or just Jira) |
How to actually turn one on
- Open your connector settings. Go to Settings, then Connectors, inside Claude on the web or the desktop app.
- Pick your app. Browse the Connectors Directory and choose a service, or click “Add custom connector” if you need to enter your own server address.
- Sign in safely. You’ll go through the same kind of sign-in (OAuth) you use for other apps — Claude never actually sees your password, and you can revoke access at any time.
- Switch it on for your chat. Enable the connector for the specific conversation using the “+” button at the bottom of the chat box.
- Just talk normally. Ask your question the way you always would. Claude decides on its own whether it needs to reach into that connected app to answer you properly.
Anything in the official Directory has been vetted by Anthropic to some degree — think of it as an app store you can generally trust. Custom connectors are different: they haven’t been verified, so only connect to servers you actually trust, since connecting one grants Claude the ability to access and potentially act within that service based on the permissions you approve.
Real, everyday use cases
Project & delivery work
Connect Jira or a similar tracker and ask “what’s still open for this sprint?” — no more tab-switching to check status before a stand-up.
Inbox & calendar
With Gmail and Calendar connected, ask Claude to find a reply, summarise a thread, or check what’s free next week.
Docs & files
Connect Google Drive and ask Claude to pull up last quarter’s report or compare two documents, without downloading anything yourself.
Team communication
With Slack connected, Claude can send messages, search channels, and draft posts on your behalf.
And you’re not limited to one connector at a time — enable several in the same conversation, and Claude works out which tool fits which part of your question, for example combining a tracker, a chat tool, and a document store for one project update.
What changes when you use them — versus when you don’t
| Without a connector | With a connector switched on |
|---|---|
| You copy-paste text from your app into the chat before Claude can help | Claude looks the information up itself, live, while you’re talking |
| Claude only knows what’s already in this conversation | Claude can pull in current, real data from your actual accounts |
| Claude can describe what you should do next | Claude can actually do it — send the message, create the ticket, update the file, where you’ve allowed it to |
| Nothing outside the chat is touched | Claude inherits your own permissions in that tool and can only see and do what you could already see and do yourself — it isn’t given extra access beyond your account |
In short: without connectors, Claude is a very capable assistant working from whatever you happen to paste in front of it. With connectors, it becomes something closer to a colleague who already has access to the same tools you do, and can go check things for you instead of asking you to fetch them first.
Turning a connector on means Claude is reading (and sometimes writing) real data in a real account. That’s the whole point, but it’s also why the sign-in step matters, and why it’s sensible to only connect apps and servers you genuinely trust — and to review the permissions each connector asks for before approving it.
The takeaway
MCP is the quiet plumbing underneath — the shared standard that lets an AI assistant and an outside app speak the same language. A Connector is the visible, clickable result of that plumbing: the specific link between Claude and Slack, or Claude and Drive, or Claude and whatever internal tool your company runs. You’ll never need to think much about MCP day to day. You will, however, get real, practical value out of switching on the right connector for how you actually work.
Enjoyed this breakdown?
ToolTechSavvy covers AI tooling, developer workflows, and practical guides like this one — written for people who actually use these tools, not just read about them. Come explore more.
Read more on ToolTechSavvy →
