Claude works with email in three different ways and none of them is an autonomous inbox. Claude for Outlook is a beta add-in for Outlook on the web, Windows and Mac: it triages your unread mail, summarises threads with citations and writes replies straight into the compose pane, but Anthropic's documentation is explicit that it never sends mail or invites on its own. The Microsoft 365 connector goes further and can send Outlook mail and create calendar events from claude.ai. The Google Workspace connector covers Gmail, where Claude can search, read and create drafts but cannot send. Crucially, none of these is event-driven: nothing happens because an email arrived. An inbox that reads incoming mail and replies on its own is still something you build, with an inbound-mail webhook in front of the Anthropic API.
Claude will draft the reply.
Sending is still on you.
Outlook and Gmail get very different deals, and neither one gives you an inbox that answers itself. Here is exactly what each can do, and what it takes to build the rest.
The short answer.
| At a glance | |
|---|---|
| Outlook | Claude for Outlook, a beta add-in on Pro, Max, Team and Enterprise. Triage, thread summaries with citations, and replies drafted into the compose pane. It never sends on its own. |
| Gmail | Reached via the Google Workspace connector. Claude can search, read, manage labels and threads, and create drafts. It cannot send email on your behalf, and it cannot read attachment contents. |
| Where Outlook goes further | The Microsoft 365 connector can send Outlook mail, manage drafts, filters and auto-replies, and create calendar events - capabilities the Gmail connector does not have. |
| Not supported in Outlook | Mobile Outlook clients, Outlook 2016 and 2019, and on-premises Exchange. PDF attachments cannot be read inline. |
| Is it event-driven? | No. Every one of these is something you start. Nothing runs because a message landed in the inbox. |
| Admin required | Claude for Outlook needs tenant-wide Microsoft Graph admin consent. Gmail may need your Google Workspace admin to approve Claude as a trusted app. |
| Build-your-own stack | An inbound-mail webhook - Postmark, SendGrid Inbound Parse or Mailgun Routes - in front of the Anthropic Messages API, replying through the same provider. |
Last reviewed 2026-07-29 View as Markdown
Your options.
Email is where Claude's capabilities are most uneven, so read this table before you promise anyone an outcome. The gap between Outlook and Gmail is real and it is not small.
| Approach | What you get | Where it stops |
|---|---|---|
| Claude for Outlook | An add-in inside Outlook itself. Triages unread mail by priority, summarises threads with per-email citations, reads .docx and .xlsx attachments inline, finds meeting times and drafts replies into the compose pane. | Beta. No mobile clients, no Outlook 2016 or 2019, no on-premises Exchange, and it cannot read PDF attachments inline. It never sends mail or invites on its own. It is also not currently covered by Enterprise audit logs. |
| The Microsoft 365 connector | From claude.ai, Claude searches your Outlook mail and can genuinely act: sending mail, managing drafts, labels, filters and auto-replies, and creating calendar events. | You are working in claude.ai rather than in your mail client, and every action is one you asked for in that moment. Write permissions need a second layer of Entra approval. |
| The Google Workspace connector | Covers Gmail alongside Calendar and Drive. Claude searches and reads mail in plain language, creates drafts in your Gmail account, and manages labels and threads. | Cannot send email on your behalf - drafts only. Cannot access attachment contents, only their metadata. Some advanced Gmail filters are unsupported. |
| An inbound-mail bot you build | The only way to get an inbox that answers by itself. An inbound-mail provider posts each message to your webhook, you call the Anthropic Messages API, and you reply through the same provider. | You own deliverability, SPF, DKIM and DMARC, threading headers, quoted-text stripping, idempotency and auto-reply loop protection. The model is the easy part of this build. |
| An AI teammate with its own inbox | A product that already has an address you can email and reply to, with the deliverability and threading solved. Beagle answers its own inbox and works the same way there as in Slack. | It is a teammate you email, not a filter sitting on your personal mailbox. If what you want is triage inside your own inbox, an add-in is the closer fit. |
Step by step.
Start with your mail platform, because it decides most of what follows. Steps one to four are the Outlook path, steps five to seven the Gmail path and the build-it-yourself path.
01 Work out which capability you actually need
Three different jobs get called "Claude in email", and they have three different answers. Triage and drafting inside your mail client is Claude for Outlook. Acting on mail while you work in claude.ai is a connector. An inbox that replies without you is a build. Naming the job first saves the wrong rollout.
02 For Outlook, get the add-in allowed and consented
An administrator has to enable Office Store access in the Microsoft 365 admin center, deploy the add-in through Integrated apps or by uploading its manifest, and grant tenant-wide Microsoft Graph admin consent. Without the consent, the add-in installs and then does nothing useful.
ScopesMail.ReadWrite, Calendars.Read, User.Read, offline_access
03 Check your Outlook clients before you announce it
Claude for Outlook runs on Outlook on the web, on Windows in both new and classic Outlook with Microsoft 365, and on Mac with a Microsoft 365 subscription. It does not run on mobile, on Outlook 2016 or 2019, or against on-premises Exchange. A field team that lives on phones will not see it.
04 Add the Microsoft 365 connector if you need it to act
The add-in drafts; the connector acts. If you want Claude to actually send an Outlook message or put a meeting in the calendar, that is the Microsoft 365 connector, and its write permissions need Global Administrator or Application Administrator approval in the Entra admin center on top of tenant consent.
05 For Gmail, connect Google Workspace and set the expectation early
The Google Workspace connector covers Gmail, Calendar and Drive. Tell people plainly that Claude will write the draft and they will press send, because Gmail drafts-only is the single most misreported fact in this whole area. On Team and Enterprise an organisation Owner enables connectors first, and your Google Workspace admin may need to approve Claude as a trusted app.
06 If you need an inbox that answers itself, put a webhook in front
Point a dedicated address at an inbound-mail provider - Postmark, SendGrid Inbound Parse or Mailgun Routes all do this - and have it POST each message to your endpoint. Verify the signature, strip the quoted history, key on the Message-ID for idempotency, enqueue the work, then call the model and reply through the same provider.
ThreadingSet In-Reply-To and References on the reply or every answer starts a new thread
07 Protect the loop before you point it at a real mailbox
A bot that replies to an out-of-office will happily talk to it forever. Check the Auto-Submitted and Precedence headers, rate-limit per sender, send from a dedicated subdomain with SPF, DKIM and DMARC aligned, and keep a kill switch within reach for the first week.
What trips people up.
Email is the channel where a small misunderstanding turns into a message a customer should never have received.
Draft is not send, and the difference is the whole product
Claude for Outlook writes into the compose pane and stops. The Gmail connector creates a draft and stops. Only the Microsoft 365 connector actually sends, and only when you ask it to in that moment. If your plan assumed replies going out automatically, none of the first-party options do that.
Outlook and Gmail are not the same product
Outlook gets a native add-in plus connector write access including send and calendar. Gmail gets read and draft. Treating "Claude and email" as one capability leads to promising a Gmail team something only Outlook can do.
Nothing here is triggered by mail arriving
Connectors and add-ins are things you invoke. There is no first-party product that watches an inbox and acts when something lands. Every "AI that handles my inbox overnight" story is either a build or a different vendor's product.
Claude for Outlook is missing from Enterprise audit logs
Anthropic documents that the Outlook add-in is not currently included in Enterprise audit logs, mailbox content leaves your tenant when using Claude accounts, and there is 30-day backend retention. For a regulated team that is a real blocker and worth raising before pilot, not after.
External senders are an injection surface
Anthropic flags prompt-injection risk from external senders in Outlook, and the same logic applies to anything you build. Mail is the one channel where a stranger writes directly into your model's context. Treat message bodies as untrusted input, and never give an email-triggered agent an unattended send.
Quoted history quietly triples your bill
In a build-your-own setup, naive parsing feeds the whole quoted chain back to the model on every reply. Ten messages in, you are paying to re-read the same conversation ten times and getting worse answers for it. Strip the quotes on the way in.
Every first-party option here stops at the draft. The interesting question is who you trust to press send, and what they read first.
Or skip the build.
Beagle takes the other side of the problem: instead of sitting inside your mailbox, it has one. You email it like a colleague, it does the work across your stack, and it replies in the thread - and it still waits for a human before anything reaches a customer.
It has an address, and it answers it
Away from Slack, email Beagle the way you would email a teammate. Deliverability, threading and the reply chain are already solved, so there is no webhook for you to run.
The reply is drafted from real records
Beagle pulls from the CRM, the tracker, the docs and the inbox before it answers, and shows what it used. It is not writing from the thread alone.
It picks work up on a schedule too
The Monday summary, the follow-up nobody sent, the account that has gone quiet. Beagle runs recurring jobs rather than waiting to be prompted.
Client-facing sends wait for approval
Anything going outside the company arrives drafted, in your DMs or your inbox, for you to approve. The one place an autonomous send would be unforgivable is the one place Beagle will not do it.
Plugged into your stack.
OAuth in, every read scoped to the teammate who asked.
Common questions.
Can Claude read and reply to my email?
Claude can read your email and write replies, but it will not send them for you in most setups. Claude for Outlook drafts into the compose pane and Anthropic states it never sends mail or invites on its own. The Gmail connector creates drafts and cannot send. The one exception is the Microsoft 365 connector, which can send Outlook mail and create calendar events when you ask it to.
Does Claude work with Gmail?
Yes, through the Google Workspace connector, which covers Gmail alongside Calendar and Drive. Claude can search and read your mail in plain language, manage labels and threads, and create drafts in your Gmail account. It cannot send email on your behalf and it cannot read the contents of attachments, only their metadata.
What is Claude for Outlook and what does it need?
It is an add-in that runs inside Outlook on the web, Windows and Mac, available in beta on Claude Pro, Max, Team and Enterprise. It triages unread mail, summarises threads with per-email citations, reads .docx and .xlsx attachments inline and drafts replies into the compose pane. Deployment needs Office Store access enabled in the Microsoft 365 admin center and tenant-wide Microsoft Graph admin consent for Mail.ReadWrite, Calendars.Read, User.Read and offline_access.
Can Claude handle my inbox automatically while I sleep?
Not with anything Anthropic ships today. Connectors and add-ins are invoked by a person; none of them is triggered by a message arriving. An inbox that reads and answers mail on its own is something you build, with an inbound-mail provider such as Postmark, SendGrid Inbound Parse or Mailgun Routes posting to a webhook that calls the Anthropic API and replies through the same provider.
Does Claude for Outlook work on mobile?
No. Claude for Outlook supports Outlook on the web, Outlook on Windows in both new and classic with Microsoft 365, and Outlook on Mac with a Microsoft 365 subscription. Mobile clients, Outlook 2016 and 2019, and on-premises Exchange are not supported.
Is it safe to give an AI access to a work inbox?
Email is the highest-risk channel to automate, because an external sender writes directly into the model's context. Anthropic itself flags prompt-injection risk from external senders. Insist on drafts rather than unattended sends for anything customer-facing, scope access to a single mailbox rather than a whole tenant, and check the audit story first - Anthropic documents that Claude for Outlook is not currently covered by Enterprise audit logs.
