How to Connect Gmail to Claude, and What It Can Do

Most people who search for this have already decided. They want the thing that reads their inbox to be the same thing they already talk to all day, instead of a separate tool with its own rules. The connection itself takes about a minute. Knowing what it will and will not do afterwards takes a bit longer, and that is the part that saves you from a bad surprise.

There is no password to hand over and no API key to paste. Google stopped supporting the old style of access where an app asks for your account username and password, and for Google Workspace accounts that route was closed on 1 May 2025. Everything now runs through a sign-in where Google, not the app, decides what gets granted.

What does connecting Gmail to Claude actually give you?

Search and read across your mailbox in plain language, plus drafting. Anthropic's help centre describes the Gmail connector as searching and reading emails using natural language queries, drafting emails with context, listing saved drafts in your Gmail account, and working with email metadata and labels.

Anthropic's connector page for Gmail goes further and describes reply, forward and send with your approval, noting that Claude asks for approval by default before anything goes out. The help centre article on Google Workspace connectors is more conservative and says Claude creates drafts in your Gmail account but cannot send. Treat the sending question as the one to check inside your own account rather than assume, because it clearly moves.

One detail matters more than it sounds. Anthropic states that Claude only accesses the Gmail, Calendar and Drive data for the Google account you connected, only when you ask a question or request an action that needs it, and that it retrieves the minimum information needed and cites the emails it used. So the connector is not a background watcher. Nothing happens in your inbox while you are asleep.

Which plans and accounts can turn it on?

Anthropic's help centre says the Google Workspace connectors are available for all users on Claude and Claude Desktop, and its Gmail connector page lists Claude, the desktop app, the mobile app, Claude Code and the API as surfaces where the connector applies. Individual accounts sign in with Google directly.

Work accounts are where people get stuck. On Team and Enterprise plans, an Owner or Primary Owner has to enable the connectors at the organisation level first. If the Gmail option is missing or greyed out for you and a colleague has it, that is the most likely reason, and it is a two minute ask rather than a support ticket.

Connecting Gmail to Claude, start to finish

1. Check the account you are signed into — Personal accounts can connect on their own. If you are on Team or Enterprise, confirm an Owner or Primary Owner has enabled Google Workspace connectors for the organisation, because individual sign-in is blocked until they have.
2. Open Claude's connector settings — Go into Claude's settings and find the connectors section. Anthropic's help centre refers to Customize, then Connectors, when you go back later to manage or remove one, so that is the area you are looking for.
3. Choose Gmail and sign in with Google — Pick the Gmail connector and sign in with the exact Google account whose mailbox you want reachable. Claude can only access data for the account you connect, so if you run a personal address and a work address, connect the one you actually live in.
4. Read Google's permission screen before you click Allow — This screen is the only place you see the full list of what is being granted, in Google's words rather than the app's. Read it. If something on it surprises you, stop there, because after Allow the grant is live.
5. Prove it works with one small question — Ask for something you can verify instantly, like the last email from a named person, or a summary of a thread you already know. If the answer cites real messages, the connection is good.
6. Decide your approval habit now, not later — If your setup allows Claude to send or reply, keep the approval prompt on and read what it wrote before it goes. The first time you skim and approve a wrong recipient is the moment you learn why.

People skip that permission screen because it looks like every other one. It is not. It is the single moment where reading takes ten seconds and not reading can take an afternoon of undoing. Whatever appears there is what the app can do until you remove it, no matter what a marketing page said.

Where exactly is the line between read, draft and send?

Reading is the safe part, drafting is the useful part, and sending is the part with consequences. A draft sits in Gmail until a human opens it. A sent message is a fact about your business that you cannot take back, which is why approval prompts exist and why turning them off is the one shortcut worth refusing.

That third row is why we built approval into the design rather than bolting it on at Sistava. An AI Employee that handles a mailbox should tell you what it is about to send, in the same place you already read its work, and wait. Speed is not worth much if you spend the time you saved apologising.

Once it is connected, the temptation is to ask for something dramatic. Do the opposite for the first day. Ask questions that have a checkable answer, so you learn how it searches and how literal you have to be, before you let it write anything anyone else will read.

• Catch me up. Summarise the thread with a named person and tell me what I owe them.
• Find the number. Pull the amount and due date out of the last invoice from a supplier.
• Chase list. Show me every message from the last month where someone asked me something and I never replied.
• Draft, do not send. Write the reply to this thread as a draft and leave it in Gmail so I can read it first.

That last one is the honest test of whether this saves you real time. If the draft needs a full rewrite, the tool does not know enough about your situation yet. If it needs one line changed, you have found something worth keeping.

The difference between a chat window with a mailbox attached and an assistant is repetition. A chat answers when you open it. An assistant knows that Monday morning means the unanswered questions get gathered, the invoices get flagged, and the two people waiting on you get chased, without you starting the conversation each time.

How do you check what has access and take it back?

In two places, and you should know both. Inside Claude, Anthropic documents going to Customize, then Connectors, clicking your Google connector, choosing Disconnect and confirming. On Google's side, your account has a page listing every app you have connected, and removing access there works even if you no longer use the app.

Google's own documentation is blunt about the limit of revoking. Once you remove a linked app's access, it can no longer reach your data, but Google notes you may need to contact the app's developer to ask them to delete data they already have. Revoking stops the future, not the past, which is a good reason to be careful about what you connect in the first place. The same logic applies to any tool you give a mailbox to, including Sistava.

What this setup will not do for you

It will not work your inbox while you are away. Anthropic is explicit that Claude accesses your Gmail data when you ask a question or request an action that needs it. There is no rule that fires when a message arrives at 3am, no queue that gets worked overnight, and no follow-up that happens because a week passed.

It also will not remember your standards unless you keep restating them. Chat is a great interface for a question and a poor one for a policy. If you want refunds under a certain amount handled one way and everything else escalated, you will be typing that rule again and again.

That gap is the whole reason a different shape of product exists. You hire an AI Employee at Sistava, connect the same mailbox, and write the standing instructions once. It works on a schedule you set, it keeps a record of what it did, and it asks before anything leaves the building.

Before you connect anything to a mailbox, it is worth spending five minutes on the security side. The questions are simple, the answers are short, and asking them once is much cheaper than finding out later which tool still has a key.

Connecting is the easy half. The hard half is that your inbox does not stop when the chat window closes, and a connector only helps in the minutes you spend inside it. Most people who wire Gmail to a chat assistant get real value from search and summaries, then quietly go back to doing the actual replying by hand.

If the work you want gone is recurring rather than occasional, the answer is not a better prompt. It is something that owns the task, runs to a schedule, holds your standards, and shows you what it did. That is a different product shape, and it is worth knowing that before you spend a month trying to make chat behave like staff.

Sources

• Claude Help Center, Use Google Workspace connectors
• Claude by Anthropic, Gmail Connector
• Google Account Help, Third-party apps and services with access to your account
• Gmail Help, Add Gmail to another email client