cobrain
Create your brainGo to your brain
Blog · 20 November 2026 · 6 min

Can Claude read my Gmail? How the Gmail connector works

Yes, Claude can read your Gmail through the Gmail connector. How to connect it, what Claude can do, when it asks before sending and what it keeps.

Yes, Claude can read your Gmail. You switch on the Gmail connector, sign in with Google, and from then on Claude can search your inbox and answer questions about your email, with links back to the messages it used. Anthropic has since added sending, replying and forwarding. Each of those asks for your approval first, at least by default.

This article covers how to connect Gmail, what Claude can do with your email, what it asks permission for, and what it doesn't keep once the chat is over. We checked every claim about Claude against Anthropic's help center and its connector docs on 20 November 2026. Where the two pages disagree, we tell you.

How Claude reads Gmail: the Gmail connector

Claude doesn't see your inbox on its own. It reaches Gmail through a connector, the same mechanism it uses for Notion, Google Drive and many other services. You ask a question that needs your email, Claude works out that it should search Gmail, and it searches.

Anthropic's help center says Claude only touches your data when you ask a question or request an action that needs it, and that it fetches the minimum it needs. The answer comes with citations, so you can open the original email and check.

If you've already read our piece on Claude and Notion, this one will feel familiar. The setup is close to the same. The risks are different, though, because an inbox can send things to other people.

How to connect Gmail to Claude

From Anthropic's docs, on claude.ai:

  1. Go to Customize > Connectors.
  2. Find Gmail and select Connect.
  3. Sign in to your Google account and grant the permissions it asks for.
  4. In a chat, select +, then Connectors, and turn Gmail on for that conversation.

Connecting Gmail on claude.ai

  1. 1Open connectorsCustomize > Connectors
  2. 2Find Gmailselect Connect
  3. 3Sign in to Googlegrant the permissions
  4. 4Turn it on in the chat+ › Connectors
Once connected, the button changes from Connect to Disconnect.

On Team and Enterprise plans, Gmail won't appear in your list until an Owner or Primary Owner enables it for the organization. If your company runs Google Workspace, you may also hit "Access blocked". In that case a Workspace admin has to mark Claude as a trusted app in the Google Admin console (under Security, API controls, third-party app access), and the change takes about 15 minutes to apply.

Which plans get it is one of the spots where Anthropic's two pages differ. The help center says the Google Workspace connectors are available to all users on Claude and Claude Desktop. The Gmail page in the connector docs lists Pro, Max, Ultra, Team and Enterprise. On the free plan, go by whether Gmail shows up in your connectors list.

What Claude can do in your Gmail

According to the help center, Claude can:

  • search and read emails, asking in plain language ("what did the accountant say about the invoice?");
  • write drafts;
  • send, reply to and forward emails;
  • read email metadata, attachment metadata included, and open attachments up to about 12 MB (in chat this needs code execution and file creation switched on);
  • work with labels and threads, and list your saved drafts.

Sending is recent: press coverage dates it to 18 August 2026. The Gmail page in Anthropic's connector docs still describes a read-only connector that can't create, send or change messages, so if your account behaves that way, that page explains why. Searching and reading are on both pages without any doubt.

What Anthropic's two pages say

Help centerConnector docs
Search and read
Send, reply, forward
Plansall usersPro and up
Checked on 20 November 2026. Sending arrived in August, so the docs page looks behind.

There are limits, too. The connector docs say Claude can't see images embedded in emails. Some advanced Gmail filters aren't supported. Google's API quotas apply, and Anthropic says performance can vary with large mailboxes.

Permissions: what Claude asks before it sends

When you sign in, Google's permission screen mentions sending email. That looks alarming, and it's accurate. Anthropic says Claude can send, but by default it asks for your explicit approval before each send, reply or forward.

There's one exception you should know about. In Auto mode, those actions can run without asking you, unless an automated check flags them as risky or your organization requires approval. Anthropic itself says the check can miss risky actions. If you want to approve every email yourself, stay in Manual or set the tool to "Always ask" where that option exists. On Team and Enterprise plans, owners decide whether members may let these actions run without asking.

Claude writesa reply or a new email
asks first
You approvedefault outside Auto mode
only then
Gmail sends it
In Auto mode the approval step can be skipped. Manual or «Always ask» keeps it.

Claude also mirrors your Google permissions. It sees the mailbox of the account you connected, and nothing you couldn't open yourself. That cuts both ways: anything you can read, Claude can read when it decides a search is relevant. We wrote a longer piece on giving an AI write access, and most of it applies to an inbox as well.

On data: Anthropic says it doesn't train its models on Gmail connector data. What Claude retrieves is stored with the chat, so deleting the chat deletes it. On consumer plans, there's a footnote. If you allow your chats to be used for training, content you copy from your email into the chat, or answers that quote it, may be used.

Where a Gmail connector stops

The connector is a good way to ask your inbox questions. It isn't a memory.

Each time you ask, Claude searches Gmail again. The data it pulled lives inside that one chat. Say you work out with Claude which supplier you picked and why, after reading a dozen threads. That conclusion stays in the conversation unless you save it somewhere. Our article on how Claude's memory works explains what does and doesn't carry over between chats.

Email is also a poor place to look things up later. A decision sits in a reply on page three of a thread, next to quoted text and signatures. Searching it every time costs context, and you get the thread, not the conclusion.

Then there are your other AIs. ChatGPT, Cursor or Codex need their own way into Gmail, if they have one. What Claude figured out from your inbox doesn't reach them.

Keep what matters from your inbox in a shared memory

Cobrain is built for the part a connector doesn't do. It's a shared memory made of Markdown notes in folders, behind a remote MCP server that Claude, ChatGPT, Claude Code, Codex and Cursor can all connect to. When an AI writes a note, the note shows who wrote it, and earlier versions can be restored.

With Gmail and Cobrain both on, you can ask Claude to read a thread and save the decision in the client's folder: a few lines, not the whole conversation. Next time, any of your AIs finds the decision without opening the inbox. Connecting Claude takes a couple of minutes.

Email thread
reads
Claudewith both connectors on
saves
Cobrainthe decision, in a few lines
Gmail stays where email lives. The note is what your AIs read next time.

Quick answers

Can Claude read my Gmail? Yes. Connect Gmail from Customize > Connectors in claude.ai, sign in to Google and turn the connector on in the chat. Claude can then search and read your email and cite the messages it used.

Can Claude send emails from my Gmail? According to Anthropic's help center, yes: it can send, reply and forward, and by default it asks for your approval before each one. In Auto mode those actions can run without asking. Anthropic's connector docs page still describes a read-only Gmail connector.

Is the Gmail connector free? Anthropic's help center says the Google Workspace connectors are available to all users. Its connector docs list Pro, Max, Ultra, Team and Enterprise for Gmail. Check your connectors list to see whether it's there.

Does Anthropic train on my email? Anthropic says it doesn't train on Gmail connector data. On consumer plans with training allowed, email content you paste into a chat, or answers that quote it, may be used.

Why can't I connect Gmail at work? On Claude Team and Enterprise an Owner has to enable Gmail first. On the Google side, a Workspace admin may need to mark Claude as a trusted app; the change takes about 15 minutes.

Free, no card needed

Start with a single note.

It is free and no card is needed. You sign in with a link sent by email, connect your AI and start building a memory that does not stay locked inside a chat.

Create your brain, freeGo to your brainA thousand notes, as many AIs as you want to connect.