How to

How to let Claude draft and send emails for you

Claude can write a reply far faster than you can, but the draft usually lives in the chat window. A custom connector puts that draft in your real mailbox and lets you send it without leaving the conversation.

Updated June 24, 20264 min read

Let Claude draft and send emails by connecting your mailbox through a custom MCP connector with write tools. Claude composes a draft inside your account, in the right thread, and sending stays behind a confirmation you approve. Drafting flows freely, while the send action waits for your go-ahead every time.

Key takeaways

A custom connector lets Claude create real drafts in your own mailbox, not just text in the chat.
Drafting and replying in-thread flow freely, while sending is gated behind a confirmation you approve.
Claude acts as you, with your account's own permissions, and never reaches beyond what you can already do.
It works for Gmail, Outlook, and other providers with an API, including shared team inboxes.

Can Claude actually send an email, or just write one?

Claude can send an email when its connector includes a send tool, not just a drafting one. Most default email connections stop at reading and summarizing. A custom connector adds the write tools that create a real draft in your mailbox and, with your approval, send it from your own address.

The draft Claude writes in a chat window is only text until something puts it in your mailbox. A custom connector closes that gap by creating an actual draft in your account, formatted and addressed, sitting in the same thread you would reply to by hand.

Sending is treated as a separate, higher-stakes action. It is its own gated tool, so a message only leaves your mailbox after you confirm it. Claude never quietly fires off an email on your behalf.

How do I let Claude draft and send emails for me?

To let Claude draft and send emails, connect your mailbox through a custom MCP connector that authenticates as your own account. Then ask Claude inside a conversation to reply to a thread or write a new message. It builds the draft, you review it, and you approve the send.

A typical flow runs entirely inside one conversation:

1You ask Claude to reply to a customer or write a new message.
2Claude reads the thread with a read tool like get_thread, then drafts a response in your voice.
3It creates that draft in your mailbox with draft_reply, on the original thread.
4You review and edit the draft, then approve. Only then does send_email run.

Because Claude works inside a conversation you start, nothing happens in the background. There is no autonomous agent watching your inbox and firing replies. The drafting and sending both sit inside the same chat where you asked for them.

Read and write tools are separate

Drafting tools like draft_reply run freely, but send_email is its own gated tool, so you can let Claude prepare everything and still hold the send.

Will Claude send an email without my approval?

No. Drafting, replying in-thread, and updating a draft can flow freely, but the send action is gated behind a confirmation you approve every time. Claude can prepare everything right up to the point of sending, and the message only goes out once you say so.

You decide where the line sits. Creating and updating drafts can be automatic, while sending and deleting stay behind a human confirmation. That keeps the fast part fast and the irreversible part deliberate.

Sending is never automatic

By default, send_email and delete always require your confirmation. Claude can prepare the message completely, but it only leaves your mailbox after you approve it.

Is it safe to give Claude access to my email?

Giving Claude email access is safe when the connector is scoped properly. It authenticates as your own account and inherits your permissions, so it can never reach mail you cannot already see. Read and write tools are separate, and sending requires a human confirmation every time.

The connector never grants more access than your own account already has. If you cannot read a mailbox, neither can Claude through the connector.

You choose which actions are automatic and which need a confirmation step, so the highest-stakes actions, sending and deleting, always keep a human in the loop.

Does this work for Gmail, Outlook, and shared inboxes?

Yes. We build connectors for Gmail and Outlook, and for any email system with an API, including shared team inboxes and support aliases. Claude reads threads and creates drafts in the same account you already use, with sending scoped to the addresses your account is allowed to send from.

Frequently asked questions

Can Claude reply to an email in the right thread?

Yes. A custom connector lets Claude read the original thread and create its reply as a draft on that same thread, with the subject line, recipients, and quoted history already in place. You review the draft in your mailbox and approve the send, so the reply lands as a proper in-thread response rather than a fresh message.

Will Claude email people without me knowing?

No. Claude works inside a conversation you start, so there is no background agent sending mail on its own. Drafting can happen automatically, but the send action is a separate gated tool that waits for your approval. Nothing leaves your mailbox until you confirm it, and you control which actions require that human step.

Which email providers can you connect Claude to?

Any provider with an API. We build connectors for Gmail and Outlook, and for other mail systems your team runs, including shared inboxes and support aliases. If you can sign in and send from an account, Claude can draft and, with your approval, send from it through a custom connector scoped to your own permissions.

Keep reading

Full read-write Claude connector for Gmail Full read-write Claude connector for Outlook Case study: Gmail in-thread drafting