Claude Can Now Send Your Emails, Not Just Draft Them

For three years, your AI assistant has been a very good intern who is not allowed to touch the keyboard.

It writes the email. You copy it. You paste it into Outlook. You check the tone one more time, because it is going to a client. You hit send yourself.

It drafts the agenda. You open your calendar. You find the slot. You type the invite. You send it.

It suggests the file update. You open SharePoint. You make the change by hand.

Priya runs a nine-person bookkeeping practice out of a small office in Toronto. Every Friday afternoon, she does a version of this loop about fifteen times: invoice reminders, meeting confirmations, a shared client tracker in OneDrive that needs updating after every call. Claude has been drafting the words for her for a year. It has never been allowed to actually do any of it.

What changed on July 7, 2026

Anthropic flipped a switch. The Microsoft 365 connector inside Claude, which used to only read and search your email, calendar, and files, can now write to them.

Before this, the honest description of Claude plus Microsoft 365 was: a very well-read assistant that could tell you what was in your inbox, but had to hand everything back to you to actually act on it. Read, search, summarize. That was the ceiling.

The ceiling is gone. Claude can now draft an email and send it. Book the meeting and put it on the calendar. Open the client tracker in SharePoint and update the row. In one instruction, not four separate trips through four separate apps.

The one-sentence reframe

You were never short on things to say to your clients. You were short on the twelve minutes it takes to actually say them, in the right app, in the right place, twelve times a day.

That twelve minutes, multiplied by every small business on Outlook, is the real cost center. Not creativity. Logistics.

Can AI send emails and manage my calendar directly inside Outlook?

Yes, and this is the part most guides to "AI scheduling assistants" miss. Search for help with this today and you mostly find third-party add-ons: Motion, Reclaim.ai, Calendar Bridge, or a custom bot wired up through Make.com. All of them ask you to connect a new tool to your existing one.

Claude's Microsoft 365 connector skips that step. It is not a bolt-on scheduler watching your calendar from the outside. It works through Microsoft's own Graph API, inside your existing Outlook, OneDrive, and SharePoint account. No second subscription, no new dashboard to check, no data leaving Microsoft's system to live in some other company's database.

A worked example: the Friday email run

Here is the before and after, in Priya's practice.

Before, with a read-only assistant: "Draft a follow-up to the Kowalski account about their overdue invoice." Claude writes a clean, polite draft. Priya copies it into a new Outlook message, checks the client's email address, adjusts one line, and sends it herself. Repeat fifteen times.

After, with write tools on: "Check the SharePoint client tracker for any invoice more than 14 days overdue, draft a follow-up email for each one in my voice, and hold them for my approval before sending." Claude reads the tracker, finds the three overdue accounts, writes three separate drafts sitting in Outlook's Drafts folder, and waits. Priya opens each one, reads it in ten seconds, and either sends it or edits a line first.

Nothing about the writing changed. What changed is that the four manual steps between "good draft" and "sent email" collapsed into one review click.

Isn't this just a chatbot sending my emails unsupervised?

Fair question, and Anthropic built the answer into the feature rather than leaving it to trust. Three guardrails matter here:

• Every tool has its own permission level. In Claude's connector settings, each capability, sending mail, deleting a calendar event, editing a SharePoint file, can be set independently to needs approval, allow once, or always allow. Nothing is all-or-nothing.
• Agent-written email is labeled. Anthropic confirmed that mail Claude sends through the connector carries a header identifying it as agent-sent, so it is never disguised as typed by you.
• An admin has to say yes first. On a Microsoft 365 work or business account, a Microsoft Entra administrator must consent to the updated permission set before anyone in the organization can use write tools. Teams chat access stays read-only regardless. On a personal Free, Pro, or Max account, you approve it yourself by reconnecting the connector.

The default posture is caution, not autonomy. You decide how much rope to give it, tool by tool.

How do I turn on Claude's Microsoft 365 write tools?

The setup splits into two paths depending on your plan.

If you are on Claude Free, Pro, or Max (a solo owner or a small team without a shared enterprise account):

1. Log into Claude on the web or desktop app.
2. Click your name in the bottom left, then Settings, then Connectors.
3. Find Microsoft 365 and click Disconnect, then Connect again. Reconnecting is what pulls in the new write permissions.
4. Sign in with your Microsoft account when prompted, and consent to the updated access.
5. Open the Microsoft 365 connector again and scroll to the read, write, and delete tool list. Everything defaults to needs approval. Set each one the way you want it.
6. Start a new chat and test it: "Send a short email to [name] confirming our meeting time." Claude will ask permission the first time, then act.

If you are on Claude Team or Enterprise (a shared organization account):

1. A Microsoft Entra global administrator needs to review and consent to the new permission set for the whole organization, from the Entra admin center, not from inside Claude.
2. If your organization set up the Microsoft 365 connector before July 7, that consent screen sometimes will not appear cleanly on the first try. If it stalls, remove the existing Claude enterprise application in Entra, wait, then reconnect from Claude's side. Do this outside business hours; the re-authentication can take a few hours to fully propagate.
3. Once reconnected, open the connector's tool list. Team and Enterprise accounts start with almost everything blocked. Turn on read, write, and delete permissions individually for what your team actually needs.
4. Test with a low-stakes prompt first: have Claude draft, not send, an email to a colleague, and confirm the draft appears correctly in Outlook before allowing direct sends.

Where this earns its keep first

Not every task belongs to an agent yet. The tasks that are ready today share one trait: they are the ones where the thinking is already done and only the typing is left.

None of this replaces judgment. It replaces the twelve minutes of moving your own judgment from one app into another.

Where to be careful

Set delete permissions to needs approval until you trust the pattern, especially for calendar events and mailbox rules; an agent that can delete is an agent that can delete the wrong thing. Start with drafts before direct sends on anything client-facing. And remember the agent-written header is a feature, not a bug: if a client ever asks whether you personally wrote something, the honest answer should already be built into the email.

This sits alongside the same broader shift we have covered before: Claude Cowork turning chat into ongoing work sessions, and agents that write the first outreach email. Write tools are the missing link between an agent that talks and an agent that finishes.

What you have after this

A week in: your Friday email run is a five-minute approval pass instead of a forty-minute typing session.

A month in: your calendar books itself from a single instruction, and your SharePoint client files stay current without you opening them every time something changes.

Six months in: the twelve minutes you used to spend moving your own words between apps has become the two minutes you spend deciding whether to say them at all.

Same inbox. Same calendar. Same files. A completely different relationship to Friday afternoon.