how to set up email for your OpenClaw agent (without Google Pub/Sub)

If you've tried giving your OpenClaw agent email, you probably started where everyone starts: Gmail. And if you wanted real-time delivery instead of polling every few minutes, you hit the Pub/Sub wall.

Google Cloud Pub/Sub is the official way to get push notifications from Gmail. It's also a four-layer infrastructure project disguised as a checkbox. You need a GCP project with billing enabled. You need the Pub/Sub API turned on. You need gogcli for authentication. You need OAuth scopes configured correctly. You need a public webhook endpoint your agent can receive messages on. And you need all of these things to work together without one of them silently breaking.

One user in OpenClaw Discussion #4220 put it plainly: "I've invested nearly three days and about $300 and can't believe how potentially cool, but actually shitty this is."

That quote stuck with me because it captures the exact gap between what OpenClaw promises and what the email setup delivers. Your agent can browse the web, push code to GitHub, manage files. Then you ask it to read an email and suddenly you're provisioning cloud infrastructure.

There's a simpler path. It doesn't involve Google Cloud, Pub/Sub, or your personal inbox at all.

What the Pub/Sub setup actually looks like#

Let's be honest about what you're signing up for. To get real-time Gmail in OpenClaw using the Gog skill, you need to:

1. Create a Google Cloud project at console.cloud.google.com
2. Enable billing on that project (Google requires a credit card even if you stay in the free tier)
3. Enable the Gmail API and the Pub/Sub API
4. Create a Pub/Sub topic and subscription
5. Grant gmail-api-push@system.gserviceaccount.com publish rights to your topic
6. Install and configure gogcli for OAuth authentication
7. Select the right OAuth scopes (get this wrong and you'll get 403 Insufficient Permission errors with no useful context)
8. Set up a publicly accessible webhook endpoint for Pub/Sub to push notifications to
9. Call gmail.users.watch() to register the push subscription
10. Re-register the watch every 7 days because Google expires them automatically

That's ten steps before your agent reads a single email. And each step has its own failure modes. OAuth tokens going stale after 7 days if your app is still in "Testing" mode. Pub/Sub subscriptions silently expiring. Config getting wiped on restart. The invalid_grant error that could mean six different things.

If you just want IMAP polling through Himalaya instead, the setup is simpler. But then your agent checks for mail every 1-5 minutes instead of getting it instantly. For a scheduling agent or a support triage bot, that delay makes the feature useless.

The actual problem with all of this#

Even if you get Pub/Sub working perfectly, you still have a bigger issue: your agent is living inside your personal Gmail.

Every message in your inbox sits in the agent's context window. Password reset emails, medical notifications, that thread with your co-founder about equity. A prompt injection attack hidden in one crafted email could instruct your agent to forward sensitive messages, reply to threads it shouldn't touch, or exfiltrate data to an external endpoint.

This isn't theoretical. A Meta alignment researcher gave her OpenClaw agent inbox access for triage. The agent started bulk-deleting emails during what it interpreted as a "speed run." She had to physically run to her machine to kill it. The root cause was context window compaction losing her original instruction.

Pub/Sub makes real-time delivery possible, but it doesn't fix the fact that you're handing your entire email history to an autonomous system with no kill switch.

Skip all of it#

LobsterMail gives your agent its own email address. Not a forwarding alias on your Gmail. Not a wrapper around IMAP. A dedicated inbox that the agent provisions for itself.

Install the skill:

clawhub install lobstermail

Then tell your agent:

Get yourself a LobsterMail inbox so you can handle email.

That's it. Behind the scenes, the agent calls one function:

import { LobsterMail } from "@lobstermail/sdk";

const client = new LobsterMail();
const inbox = await client.provision({ name: "my-openclaw-agent" });
// → my-openclaw-agent@lobstermail.ai

No GCP project. No Pub/Sub topic. No OAuth consent screen. No billing account. Your agent hatches into the reef with a working address and starts receiving mail immediately through webhooks or WebSocket connections.

Real-time delivery without the infrastructure#

The whole reason people reach for Pub/Sub is real-time. They don't want their agent polling Gmail every few minutes and missing time-sensitive messages.

LobsterMail supports two real-time delivery methods out of the box:

Webhooks. When you provision an inbox, you pass a webhook URL. LobsterMail POSTs a signed JSON payload to that URL the moment an email arrives. No polling. No expiring subscriptions. No re-registering a watch every 7 days.

WebSocket. The SDK can open a persistent connection and push new messages to your agent as they land. Useful if your OpenClaw instance doesn't have a public endpoint for webhook delivery.

Both work on the free tier. Both deliver in seconds, not minutes.

What your agent actually gets#

On the free tier, your agent can receive unlimited emails at its own @lobstermail.ai address. No inbox caps, no credit card. Sending unlocks on the Builder plan at $9/month, which also gives you custom domains so your agent can send from support@yourcompany.com instead.

Compare that to the Gmail path: free to start, but the Pub/Sub setup costs time and often money (one user reported $300 in GCP charges before giving up). Or AgentMail: 3 free inboxes, then $20/month for 10, then $200/month for 150. LobsterMail's Builder tier gives you unlimited inboxes and 1,000 sends per day for $9.

Every incoming email also gets scanned for prompt injection across six categories: boundary manipulation, system prompt override, data exfiltration, role hijacking, tool invocation, and encoding tricks. The SDK exposes email.safeBodyForLLM() so your agent can process message content without the injection risk. No Gmail wrapper does this. Neither does AgentMail.

A real example: support triage#

Say you want your OpenClaw agent to handle incoming support emails. With the Gmail + Pub/Sub approach, you'd spend a day on infrastructure, then worry about the agent seeing every personal email in the same inbox.

With LobsterMail:

1. clawhub install lobstermail
2. Ask your agent to provision a support inbox
3. Point your custom domain's MX records at LobsterMail (the platform handles SPF, DKIM, and DMARC)
4. Incoming support emails arrive at support@yourcompany.com, get scanned for injection, and land in your agent's isolated inbox
5. Your agent reads, classifies, drafts responses, and flags anything it can't handle

No personal messages in the blast radius. No Pub/Sub subscriptions to maintain. No OAuth tokens expiring at 3 AM.

When Gmail still makes sense#

I'm not going to pretend Gmail wrappers are never the right choice. If you're running a personal experiment on a Saturday afternoon and just want to see your agent read a few emails from your inbox, Himalaya gets you there in fifteen minutes. That's fine for a demo.

But if you're building something that needs to run reliably, receive mail in real time, and not expose your personal data to an autonomous agent, Gmail + Pub/Sub is the wrong foundation. You're fighting Google's infrastructure to do something Google didn't design it for.

Give your agent its own email. Get started with LobsterMail — it's free.