Claude can now send Gmail, not just draft it. What changed in August 2026, what Claude still cannot do with your inbox, and the two settings to check before you connect. The safe setup, step by step.
Key takeaways
- The connection is official and takes minutes: Claude's built-in Gmail connection links one Google account and works across the web app, desktop, and Claude Code once enabled.
- New in August 2026: Claude can now send, reply to, and forward Gmail on your behalf. By default it shows you the message and asks approval before anything goes out; on Team and Enterprise plans, owners can allow sending to run without the per-message prompt.
- Check two things before connecting: the approval behavior for actions, which now includes sending, and, on individual plans, the data-training setting, so you know exactly what you have agreed to.
Table of contents
Connecting Gmail to Claude is genuinely simple: it is a built-in connection, set up from Claude's settings in about three clicks, and once linked Claude can search your mail, summarize threads, and write replies. And as of August 2026, one big thing changed: Claude can now send. Ask it to reply to a thread and it drafts the response, shows it to you, and on your approval sends it from your address. That makes the two settings this guide has always told you to check first matter more, not less: what Claude can do without asking you, and what happens to your conversation data. This guide covers what changed, the setup, and both checks.
This guide covers Claude's built-in route end to end. If you are weighing it against Google's new official Gmail MCP server, community servers, or a shared-inbox connection, our Gmail MCP comparison maps all five routes side by side.
What changed in August 2026
Until this month, Claude's Gmail connection was a reader and a drafter: replies it wrote landed in your drafts folder for you to send yourself. Anthropic has now shipped sending. Claude can reply to threads, forward messages, and compose and send new mail from your account. The control that makes this sane is approval: by default, Claude shows you the exact message and asks before it sends, every time. On Team and Enterprise plans, organization owners decide whether members can let those actions run automatically without the per-message prompt.
Our honest read: for a personal inbox this is a real upgrade, and the approval prompt is the right default to keep. For a shared address like support@, it raises the stakes rather than solving the problem, and the section on team inboxes below explains why.
What Claude can and cannot do with your Gmail
Worth knowing before you grant anything, because the boundaries are the point.
| What Claude can do | What Claude cannot do |
|---|---|
| Search your mail with plain-language questions | It cannot read what is inside attachments: it sees that they exist, and their names, but not their contents |
| Read and summarize threads, and surface things you have forgotten to answer | It connects to one Google account at a time, so a personal-plus-work setup needs a deliberate choice, which our multiple-accounts guide covers in full |
| Write replies with the context of the conversation | It cannot see or work a shared team address as a team: connected through your login, it acts as you, invisibly to everyone else |
| Send, reply to, and forward email from your address (new, August 2026), asking your approval first by default | It does not bypass Google's limits: sending and reading run inside Gmail's own API quotas |
The two settings to check first
- Check the approval behavior for actions.
- Check Model Improvement in Privacy settings.
- Connect.
First, approval. Claude asks before it takes actions on your connected apps, and now that sending is an action, this setting is the difference between an assistant that proposes and an assistant that acts. Leave the per-message approval on while you build trust: it turns every surprise into a question instead of a sent email. If you are on a Team or Enterprise plan, know that your organization's owner controls whether auto-send can be enabled at all.
Second, on individual plans, the data setting. On the Free, Pro, and Max plans a setting called Model Improvement, found in your Privacy settings, controls whether your conversations with Claude are used to help improve Anthropic's models. Once Gmail is connected those conversations can include the contents of the emails Claude reads, so if your inbox touches anything sensitive, check this before connecting, not after.
Neither check takes more than a minute, and doing them first is the whole difference between a safe setup and a hopeful one.
Drag AI
The inbox your team and your AI work in together
Shared inbox, live chat, and AI in Gmail, with an MCP server your AI tools can drive.
Connecting Gmail to Claude, step by step
Gmail is a built-in connection, available on Claude's paid plans: Pro, Max, Team, and Enterprise.
In Claude's settings, find the connectors section, choose Gmail, and select connect. You will be handed to Google's own sign-in, where you choose the account and review the permissions Claude is requesting, and approving returns you to Claude with the connection live. From there, just ask something your inbox knows: what needs a reply from this week, or a summary of the longest thread you have been avoiding.
If you use Claude in more than one place, the connection travels: once enabled, it applies across the web app and desktop, and Claude Code picks it up automatically. One practical note: sign in through the app's own flow, not a terminal, since Google only accepts the official sign-in route. And if your day happens in a code editor rather than a chat window, the same job is done through MCP: our Connect Gmail to Cursor guide covers it. And if your team runs on Google's own assistant, our Connect Gmail to Gemini guide covers both directions of that setup.
What to ask it first
For a deeper bench than these starters, the Drag prompt library collects our prompt collections for email, triage, and support, all of which work over this connection.
The connection is only as useful as what you ask. Six prompts worth trying in your first ten minutes, all read-only, so nothing can go out while you are still calibrating trust:
Run this directly from Claude with Drag
ClaudeClaudeClaudeClaudeClaudeAnd when you are ready to let it act, the seventh prompt is the new capability in one line:
ClaudeWith approval on, Claude shows you the reply and waits for your yes before it sends. When seven prompts stop being enough, our Claude prompts for email guide has fifteen more, each one annotated with why it works.
What about a work or team inbox?
Two honest caveats for work use. On Team and Enterprise plans, an administrator has to enable connectors before you can, so if the option is missing, that is why. And connecting a shared address like support@ through your personal login was already the wrong tool for the job when Claude could only read; now that it can send, it is the wrong tool with higher stakes. The AI would be sending as you, from a team address, invisibly to your team: no assignment, no record of who approved what, no way for a teammate to see that an answer already went out. Shared addresses belong in a shared inbox with its own AI connection, one the whole team can see and govern. That is what Drag, our shared inbox for Gmail, does: it costs $12 to $24 a seat, with AI included from $18 and nothing metered, after a 7-day trial with no card, and the setup guide for connecting it to Claude takes minutes. For the wider picture of what else Claude can reach, the best Claude MCP servers for business audits the options, and the no-code connector guide covers setup and first prompts for each one.
Frequently asked questions
Is it safe to connect Gmail to Claude?
The connection itself is official and uses Google's own sign-in, so no passwords are shared. Safety in practice comes from the approval setting: Claude asks before it sends, replies, or forwards, and you can keep that per-message prompt on permanently. Check the approval and data settings before connecting and the setup is as safe as this category currently gets.
Can Claude send emails from my Gmail?
Yes, as of August 2026. Claude can send, reply to, and forward email from your connected Gmail account. By default it shows you the message and asks your approval before anything goes out; on Team and Enterprise plans, organization owners can allow members to run sending without the per-message prompt.
Can Claude send email automatically, without asking me each time?
Only if that is deliberately enabled. The default is approval per message. On Team and Enterprise plans, the organization owner controls whether automatic sending can be switched on at all, which is the right place for that decision to live.
Does Claude read my attachments?
It can see that attachments exist and their names, but not their contents. If you need Claude to work on an attachment, share the file with it directly in the conversation instead.
Can Claude work on a shared team inbox?
Not through the personal Gmail connection, which acts as one person, and now sends as one person, invisibly to the team. Teams do it through MCP: Drag, our shared inbox for Gmail, exposes its own MCP server, so Claude can read, triage, draft, and reply on the shared queue with every action visible to the team. Drag costs $12 to $24 a seat, with AI included from $18.
Can I connect more than one Gmail account?
Not simultaneously through the built-in connection, which holds one Google account at a time. The honest routes for several accounts, including what teams should do instead, are covered in our multiple-accounts guide.
Co-founder
Building Drag for nearly ten years: shared inboxes, boards, and now the AI and agent layer, all on Gmail, plus HeyHelp for the personal inbox. Writes the honest versions of the comparisons.
