Skip to content
EDGECASTS
← Back to library
8:30 · Jun 13, 2026

Connect Claude to Your Gmail

Connect Claude to Gmail and stop being the courier between your inbox and AI. Jason walks through setup, permissions, and a draft-first workflow where Claude prepares and you pull the trigger.

Full transcript +
We've talked about how connectors in Claude work, and really, they're doors. They're doorways to a completely different part of the world. Most of the time, this grants Claude access to the internet or to particular applications that you have access to, that you need your assistant to have access to. We give access to our calendars or our mailboxes to virtual assistants across the world all the time, and doing so with Claude is actually much safer, in my opinion. I'll show you some of the reasons why as we go through this video.The main thing I want to talk about is how we can leverage these tools for the highest amount of value in your business, and probably the highest-leverage door is email. By the end of this video, Claude will be able to read your email.Your inbox is where decisions pile up. It's the tool you touch the most, and without access to your actual work, Claude is only as good as its generic knowledge: helpful, but lacking detail, and everything requires copy and paste. You've been the courier between your inbox and AI. Today, that courier retires.Claude creates drafts in your inbox. It does not send email on your behalf. It prepares, you update, and you pull the trigger. That's the right design, and you'll see why.First, we're going to select Customize. If you're on the desktop application, that'll be under Cowork. We go to Customize, then Connectors, click the plus sign, and select Browse Connectors. Here we see a whole litany of connectors to choose from. You could connect to Gmail, or you could search for Microsoft and find Microsoft 365 if you want to get into your Outlook or Microsoft 365 email. We're going to use Gmail. These are official Anthropic and partner connectors, the ones that come built in, and we're going to connect this one.When we connect, we can see the tools: create drafts, get thread, list drafts, list labels, and search threads. This particular connector is developed by Google, so this is an official Gmail-to-Claude connector that will draft replies (draft replies, not send them), summarize threads, and search your inbox. These are things I'm comfortable having it do. I'm glad it's done through Google, so I'm going to connect.When we click Connect, I sign in with a particular email address and click the account I'm signed in with. It says "Sign in to Claude for Gmail," just like any other OAuth flow you've gone through, and we look at what it can access. It can view your email messages and settings. Okay. It can manage drafts. It says it can send emails, but we saw that Claude won't be doing that. If you're not comfortable with that permission, you don't have to check it. "Read, compose, and send emails from your Gmail": again, you don't have to check it, but Claude has told us, through its official plugin from Google, that it will only make drafts. I'm going to grant it, because I need it to read, and I don't mind if it composes. I'm just not going to let it send. We'll also delve into the granular details.Now that the Gmail connector is connected, we can look at each permission. The read-only tool that retrieves a specific email thread from an authenticated user's Gmail is set to "always allow." I could say no, I want approval any time you read an email thread, or I could block it entirely. But I'll let it read. List user labels: I'll allow that too, but you could require approval or block it. Search email threads: I'm good with that as well, but you can set it to needs approval.The things that are set to need approval by default are the things that can write: creates a new draft email, creates a new label, deletes a label, adds a label to a message, adds a label to a thread, lists draft emails, removes labels from a message, removes labels from a thread, modifies a label. Notice none of those is "send a message." I have all of mine set to needs approval. You could set them to blocked, or configure each one however you like. It's really nice that we get that kind of granularity.Now let's interact with it. I go back to a chat, and it says, "Good evening, Jason at EdgeCasts." I ask it to summarize my unread emails from the last 24 hours in three buckets: needs my reply today, can wait, or ignore. I'm searching my jdstraughan account, not my EdgeCasts account, so this is from my coaching business, my main practice.It comes back: here are your unread emails. Nothing needs a reply. There's a security alert for a new sign-in on my Mac that I should take a look at, and a DMARC report, which is one of those automated checks that make sure my email is safe. "Your inbox is looking clean. The only one worth a moment of attention is the Google security alert, just to make sure that Mac sign-in was you." And it was indeed me. So we can see it's connected and working.A few principles. Every active connector carries a tool description that Claude has. Go look at those and make sure you're only using connectors you need, built by the people you trust. Have a draft-first workflow when it comes to Gmail: it creates drafts, you review, you send. And know the limits. Let it access the metadata, but don't expect it to just reply to every email for you, because it's not allowed to. That's by design.If you're thinking about confidentiality, and you should be, look ahead to future videos, because I'll be talking about how to set your settings to ensure your data isn't shared with Anthropic or used for training. Be sure to check out that video on security settings in Claude.You gave your EA access to your inbox because you wanted them to leverage it, and you knew it was worth it. You trusted that the controls were in place. Think about the calculus here. Not only do we have better controls than when I handed my inbox to somebody across the world, I can control it on a granular level. And in future videos, I'll show you how to log every single action Claude takes, so you can go back and see everything it ever did.In future videos, we'll talk about adding calendars, putting those two connectors together, creating to-do lists and daily briefs from them. We'll also look at skills, custom skills, and scheduled tasks.So go play with some connectors. Add a few more. Do a couple of tasks where you're talking to and from your inbox, from Claude, from Cowork, from a chat. And I'll see you in the next video.

Your inbox is where decisions pile up, and until now you've been copy-pasting between it and AI. In this episode, Jason connects Claude to Gmail using the official Google connector and shows the whole setup start to finish: browsing connectors, the OAuth sign-in, and the granular permission controls that let you decide what Claude can read, what needs your approval, and what's blocked entirely.
The core design principle: Claude drafts, you send. The connector can read, search, summarize, and prepare replies, but it cannot send email on your behalf. Jason demos a real inbox triage, sorting unread mail into needs-reply, can-wait, and ignore, then covers the trust calculus: compared to handing your inbox to a remote virtual assistant, Claude gives you tighter, per-tool controls and (covered in a future episode) a full log of every action it takes.