Connect Claude to your email so it can read, draft and send for you

Yes, Claude can work inside your actual inbox. Not a copy of it, not a description of it. Once connected, it reads your real emails, drafts replies in your voice, and sends them from your own address, with your permission at every step.

Most people never get this far. They use Claude in a browser tab, paste an email in, copy the reply back out, and paste it into Gmail. That works, and it is also the slowest possible version of what is available.

Here is what the real connection is, what you need before you start, and the one step that catches almost everybody.

What the connection actually is

The link is called MCP. It is a direct, permissioned connection between Claude and your real accounts.

The difference matters more than the acronym. Without it, Claude can only discuss the email you paste in. With it, Claude opens your Gmail, your Calendar and your Drive itself, and does the work inside them.

Connecting your Google account switches on all three at once:

• Gmail. Read, search, draft and send from your own address.
• Calendar. See what is on, move things, book things in.
• Drive. Open, read and edit your real files.

That is the shift from a chatbot that advises you to something that does the task.

What you need first

• A Mac. The self-paced course is Mac only. If you are on a PC, the alternative is a virtual install run live, which is scoped separately.
• A Claude subscription, from around AUD $34 per month, paid to Anthropic. This is bought separately and is not part of any course fee.
• Claude Code, running in the Claude Desktop app rather than a browser tab. Browser Claude cannot hold this connection.
• Roughly half an hour, and your Google login to hand.

You do not need to know how to code. You do not need to understand what MCP stands for. You will spend most of the time clicking through screens that Claude tells you to click.

How the setup goes

You paste one prompt into a Claude Code session, and Claude walks you through the whole thing, one step at a time. It guides every click, makes the connection, then tests it for you.

Partway through, two things surprise people, and neither is a problem:

Claude may ask you to install Node.js. That is completely normal. Follow its prompts, install it from nodejs.org, carry on.

You will create a Google app in Cloud Console. This sounds far more technical than it is. It is the standard way Google lets a tool you trust talk to your account, and Claude walks you through every screen. You are not writing anything, you are approving a connection between two things you already own.

The step that catches almost everyone

When Claude tells you to restart, you have to fully quit it with Cmd+Q.

Closing the window with the red dot is not the same thing. The app keeps running, the connection does not reload, and the Google sign-in link never appears. People sit waiting for a screen that is never going to load, and conclude the whole thing is broken.

It is not broken. Press Cmd+Q, reopen Claude, and the sign-in link appears.

If you take one thing from this page, take that.

If the sign-in link still does not appear

Three fixes solve almost every remaining case. You do not need to understand them. Paste any of them to Claude and it will sort it out.

1. Check you fully quit with Cmd+Q, rather than closing the window. This is still the most common cause on the second attempt.
2. Ask Claude to use the full path to npx, by having it run which npx, and to add the --yes flag.
3. Ask Claude to trigger the sign-in link manually.

That is the entire troubleshooting list. Between the restart and these three, there is very little left that goes wrong.

On Outlook

Microsoft Outlook connects too, by the same shape of process. You paste a prompt, Claude walks you through creating a Microsoft app, then connects and tests it.

The Cmd+Q rule applies there as well. The one extra thing to watch is that the environment variable names have to be exactly MS365_MCP_CLIENT_ID and MS365_MCP_TENANT_ID. A near-miss on either produces a connection that looks fine and never signs in.

Pick one email account and connect that. The one you actually use for work, not all of them.

Is this safe

This is the question people ask before anything else, so here is the direct answer: you decide exactly what it can reach, and you can withdraw it at any time.

The connection runs through Google’s own permission system, the same one you use every time you let an app see your calendar. Nothing is copied anywhere. Claude is reaching into your account with your permission, in the moment, and that permission is revocable from your Google account whenever you want.

Connect one account first. See what it does for a week. Add the rest when you are comfortable.

What this looks like once it is working

The point of connecting your email is not the connection. It is what stops being your job afterwards.

• Your inbox triaged before you open it, with the things that actually need you separated from the noise.
• Replies drafted in your own voice, sitting in your drafts folder, ready to check and send.
• Long threads summarised down to what was agreed and what you owe someone.
• The follow-up you meant to send on Tuesday actually sent on Tuesday.

None of that requires a new app to learn. It happens in the Gmail you already use.

---

This is Step 10 of Saviwith Foundations. The course gives you the exact prompt to paste, the full Cloud Console walkthrough screen by screen, and a recording of the whole setup being done start to finish, plus the twelve other steps that get you from a fresh Mac to an AI that knows your business.

Thirteen steps, at your own pace, lifetime access.

Start your course now · $497