How to
How to connect Claude to Gmail with full write access
Anthropic ships a native Gmail connection, and it genuinely covers reading: search, threads, summaries, even drafts. The question is what happens when you want Claude to reply inside the thread, run your labels, and actually send.
Connect Claude to Gmail either with the native connection, which searches and reads mail and can prepare drafts, or with a custom MCP connector when you need full write access. A custom build adds in-thread reply drafting, label management, archiving, and gated send, authenticated as your own Google account through OAuth.
Key takeaways
- Claude has a native Gmail connection that reads well: it searches mail, reads threads, and can prepare drafts, but it does not send.
- The gaps are in-thread reply drafting with recipients and quoted history intact, label triage, archiving, and send.
- A custom Gmail connector covers those with write tools like `draft_reply`, `label_thread`, `archive_thread`, and a gated `send_message`.
- It authenticates as your Google account through OAuth with the narrowest scopes the work needs, and outbound send always waits for your approval.
Does Claude have a built-in Gmail connector?
Yes. Claude ships a native Gmail connection, and it is genuinely useful for reading: Claude can search your mail, read threads, summarize your inbox, and prepare drafts. What it leaves out is the write side: replying inside the original thread, managing labels, archiving, and sending on your behalf.
Start with the honest version: if what you want from Claude is inbox intelligence, finding the thread, summarizing the back-and-forth, sketching a reply you will paste yourself, the native connection may be all you need. Connect it and try it first.
The friction shows up the moment Claude's output has to land back in Gmail. The default connection is read-leaning: in-thread reply drafting, label management, and send are limited or unavailable. So the reply Claude wrote still gets copy-pasted into a compose window by you, and the triage still happens by hand. That gap between reading and acting is what a custom Gmail connector closes.
The wider pattern holds beyond this one tool: in our June 2026 review of 264 business tools, 77 percent had no Claude connector at all, and none shipped full read-write out of the box.
How do I set up a custom Gmail connector for Claude?
A custom Gmail connector is built for you as a small MCP server that authenticates with OAuth 2.0 using the narrowest scopes the work needs. Read tools and write tools are separated, Claude acts as your own Google account, and outbound actions like send run only after you confirm them.
The path from request to working connector:
- 1Describe the Gmail work Claude should handle: the triage rules, the label system, and which messages it should draft replies for.
- 2We build a small MCP server with read tools (search, read thread) separated from write tools (draft, label, send).
- 3You authenticate through OAuth 2.0 as your own Google account, so Claude inherits your existing permissions and nothing more.
- 4Claude starts working your inbox in conversation, and outbound actions like send run only after you approve them.
Access is revocable at any time from your Google account, and the connector never holds broader scopes than the work requires.
What can Claude do in Gmail with write access?
With full read-write access, Claude drafts a reply inside the original thread with recipients and quoted history preserved, creates and applies labels to triage the inbox, archives and marks threads read as it clears them, and sends a drafted message once you approve, all inside the mailbox you already use.
The tool set maps to the inbox routine most people do by hand:
draft_replyto compose inside an existing thread, preserving recipients and quoted history.label_threadto create, apply, and remove labels as it triages.archive_threadto archive or mark threads read as the inbox clears.send_messageto send a drafted message after you approve it.
Claude can also find every thread matching a query and propose batched responses, which turns an afternoon of one-at-a-time replies into a single review pass.
Will Claude send email from my Gmail without asking?
No. Send is a separate, gated write action. Claude can read, draft, label, and archive as it works, but a message leaves your account only after you approve it. You decide which actions run automatically and which require a confirmation step, and the line is yours to move at any time.
In practice, most teams let search, reading, drafting, and labeling flow freely and reserve confirmations for send and anything batch-sized. Nothing leaves your account by surprise.
Claude only has your access
The connector authenticates as your account through OAuth and inherits your existing permissions. It never grants Claude more access to a mailbox than you already have, and you can revoke it at any time from your Google account.
When is the native Gmail connection enough, and when is it not?
The native connection is enough when you want Claude to read: search mail, summarize threads, and sketch drafts you will handle yourself. It stops being enough when the output needs to land back in Gmail: replies inside the thread, label-driven triage, archiving, and approved send. Those are custom-connector territory.
A quick self-test: watch what you do right after Claude answers a Gmail question. If you read the summary and move on, the native connection is fine. If you copy Claude's reply into a compose window, hunt down the thread, fix the recipients, and hit send yourself, you are doing the write half of the job manually, and a purpose-built connector will do it in place with your approval as the only remaining step.