Connecting Claude to your inbox (Gmail, Outlook): the full guide

Claude is formidable at reading an email thread and turning it into something usable: a summary, a decision, a draft reply. It just needs access to the mailbox — otherwise you spend your day copy-pasting. That is exactly what connectors solve: authorise once, and Claude can query Gmail or Outlook during your conversations.

What you can do once connected

The payoff is not theoretical. The uses that justify the connection in week one are almost always the same:

• Picking a deal back up after three weeks away: “summarise where the exchange with this customer stands”
• Finding a fact buried in a forty-message thread without knowing the exact keywords
• Preparing for a meeting from recent exchanges with the attendees
• Triaging the morning: what needs an answer, what can wait until tomorrow
• Pulling the binding parts out of a long exchange: dates, amounts, terms accepted

Three prerequisites

• A Claude plan that includes connectors. Plans change regularly — the quickest check is to open your settings and see whether the section exists.
• The right mail account. Google Workspace / Gmail, or Microsoft 365 / Outlook. Take care not to connect your personal account out of habit.
• Your IT department's approval if you are at a company. Many organisations block third-party apps on Google Workspace or Microsoft 365 by default. When a connection fails with no clear explanation, that is nearly always why.

Connecting Claude to Gmail or Outlook, step by step

1. 1Open Claude's settings

   In claude.ai, go to Settings, then the connectors section (labelled “Connectors” or “Integrations” depending on the version).

2. 2Pick Gmail or Microsoft 365

   Outlook goes through the Microsoft 365 connector, which also covers the calendar. Select the one matching your mailbox and start the connection.

3. 3Read the consent screen before approving

   Google or Microsoft states exactly what you are granting: reading messages, sometimes sending and managing labels. This is the one moment you decide the scope — take the thirty seconds.

4. 4Check the connection answers

   Back in a conversation, ask: “Summarise the unread emails since yesterday and tell me which ones need a reply.” If Claude actually queries the mailbox, you are live.

5. 5Start read-only

   Until you have judged quality on your own exchanges, have it prepare drafts rather than allowing sending. Trust is built on real cases, not on a demo.

The first prompts that actually work

An active connection is not enough: answer quality depends heavily on how precisely you ask. Three formulations that beat “summarise my email” by a distance:

• “In the thread with this customer, list only what we committed to, with the dates we announced.”
• “Sort my unread email into three groups: reply needed today, reply this week, no action. Justify each in one line.”
• “Draft a reply to this message matching the tone of my previous replies in this thread. Do not promise any date I have not already given.”

That last instruction deserves to be systematic: without it, an assistant tends to fill the gaps with commitments you never made.

Where Claude stops

These limits are not teething problems: they follow from the fact that a conversational assistant answers when asked. Worth knowing before you build a process on top.

• Nothing triggers by itself. An email that landed at 10pm waits for you to open a conversation and think to mention it. The assistant does not watch the mailbox.
• The connection is personal. Everyone connects their own mailbox. Shared addresses — contact@, support@, sales@ — go unprocessed, even though they usually carry most of the inbound volume.
• Business context is missing. Claude sees the email, not your CRM, your live contracts, or what this customer already bought.
• Nothing compounds. Your handling instructions are retyped every conversation, and the work leaves no trace the team can reuse.
• Writing stops at the mailbox. Even well connected, the assistant updates none of your other tools.

While you handle email one message at a time, an assistant is enough. The moment it becomes a flow, you need something else.

When volume outgrows the assistant: the Periscorp approach

Periscorp takes the problem from the other end. You do not connect an AI to your inbox to ask it questions: you connect your inbox to the platform, and each incoming message triggers an agent working with your company's context.

1. 1Connect the mailbox, shared ones included

   Gmail or Outlook over OAuth, contact@ and support@ included. Sync runs continuously, unattended.

2. 2Describe the handling you expect

   In plain language, as you would to a new colleague: qualify the request, identify the customer, prepare a reply, create the opportunity if it is a quote request.

3. 3Let the agent work when the message lands

   It reads, finds the customer in your data, gathers the history, drafts and updates the relevant records — at 10pm as at 9am.

4. 4Keep control where it matters

   You decide what goes out automatically and what needs human validation. Every action stays auditable, with its reasoning and sources.

Frequently asked questions

Can Claude send email on my behalf?

It depends on the permissions granted to the connection: some include sending. In practice, stay on drafts until you have checked quality on a dozen real cases.

What exactly does Claude see in my mailbox?

Whatever Google's or Microsoft's consent screen describes, no more and no less. That is the document to read — not the connector's marketing page.

How do I revoke access?

From Claude's settings, and above all on the provider side: in your Google or Microsoft account permissions, where revocation is immediate and authoritative.

What if I use ChatGPT rather than Claude?

The approach is close but the screens and the limits differ: we wrote a separate guide for it, linked below.

Connecting Claude to Gmail or Outlook takes ten minutes and pays off on day one. If your problem is comfort, stop there. If it is volume, the question is no longer “which AI should I plug into my inbox” but “who handles my email while I work on something else”.