how manus uses computer use to send and receive email

Manus recently shipped two features that let its AI agent interact with email: Mail Manus (forwarding emails to a bot address to trigger tasks) and My Computer (remotely controlling your desktop to open Gmail, find files, and send messages on your behalf). Both are interesting. Both come with tradeoffs that the official documentation doesn't spend much time on.

I spent a few days testing both features, reading through the docs, and poking at the edges. Here's what actually works, what doesn't, and where things get complicated.

What is Mail Manus?#

Mail Manus gives you a unique email address ending in @manus.bot. You pick a prefix (like yourname@manus.bot), and any email you forward to that address becomes a task for the Manus agent. Forward a newsletter and ask it to summarize. Forward a pitch deck and ask for a competitive analysis. Forward an email thread and ask for a draft reply.

The flow is simple: you send an email to your Manus bot address with instructions in the subject line or body, attach whatever context the agent needs, and Manus processes it as a task. Results come back to your email.

This works with any email provider. Gmail, Outlook, Apple Mail, ProtonMail. If it can send an email, it can trigger Manus. No API integration, no OAuth flow, no app switching. Just forward and wait.

The best use cases I've seen: summarizing long email threads, extracting action items from meeting recap emails, and analyzing attachments like PDFs or spreadsheets. It handles these well because the input (email content) and output (text response) are both naturally suited to email as a medium.

What is the My Computer feature in Manus?#

My Computer is where things get more ambitious. You install Manus Desktop on your Mac or Windows machine, keep it running, and the agent can remotely access your computer from any device. Think of it as giving Manus a VNC session to your desktop that it controls autonomously.

The email angle: Manus can open your Gmail (or any email client), compose messages, attach files from your local storage, and send them. The canonical example from their blog is telling Manus to "find the quarterly report on my desktop and email it to my client," and the agent navigates your file system, locates the document, opens Gmail, composes the email, attaches the file, and hits send.

This is genuine computer use, the same paradigm that Anthropic and OpenAI have been building toward. Manus just ships it as a product feature rather than a research demo.

How to use Manus to send emails via My Computer#

The My Computer feature lets Manus remotely control your desktop to compose and send emails through your existing email client. Here's how to set it up:

1. Install Manus Desktop on your Mac or Windows machine
2. Keep the Manus Desktop app running with your computer powered on
3. Connect your Gmail account in Manus Settings
4. Open Manus from any device (phone, laptop, tablet)
5. Describe the task, such as "find the Q1 report and email it to sarah@example.com"
6. Manus remotely navigates your desktop, locates the file, and sends the email

Each step happens autonomously. You assign the task and Manus handles the file search, email composition, attachment, and delivery without further input.

How to use Mail Manus to process emails#

Setting up the forwarding-based workflow is even simpler:

1. Log into your Manus account and navigate to the Mail Manus settings
2. Choose a unique prefix for your @manus.bot address
3. Add yourself (and any other senders you want) to the approved senders list
4. Forward any email to your Manus bot address with instructions in the subject line
5. Wait for the results, which arrive as a reply to your original email

You can also configure approved senders so only specific email addresses can trigger tasks. This prevents random people from consuming your Manus credits by emailing your bot address.

What Manus can and can't do with email#

Manus handles inbound email processing well. Summarization, extraction, analysis of forwarded content. These are tasks where the agent reads something and produces text output. Straightforward.

Sending emails through My Computer works, but it's indirect. The agent is literally clicking through a GUI. It opens Chrome, navigates to Gmail, clicks "Compose," types the recipient, pastes the body, and clicks send. This means it's subject to all the fragility of screen-based automation: UI changes break workflows, slow internet makes the agent wait, and any popup or notification can throw off the sequence.

There are things Manus doesn't do at all. It can't programmatically send email through an API. It can't provision new email addresses. It can't monitor an inbox in real time and react to incoming messages as they arrive. If your use case requires an agent that owns its own email address and can send and receive without a human forwarding messages or a desktop app running, Manus isn't built for that.

Manus also can't reply to emails on your behalf through Mail Manus. The bot address is receive-only for task triggering. If you want Manus to send a reply, you'd need to use the My Computer feature with your desktop running and your email client open.

The security question nobody is answering#

Here's the part that Manus documentation skips almost entirely: what happens to the emails you forward?

When you forward a client email, a financial report, or an internal thread to @manus.bot, that content is processed by Manus's servers. The approved senders feature limits who can trigger tasks, but it doesn't address what happens to the email content itself. How long is it stored? Is it used for training? Who at Manus can access it? The docs don't say.

The My Computer feature raises a different concern. You're giving an AI agent remote access to your desktop. It can see your files, open your browser, access your email client. Manus says the connection is encrypted and the agent only acts on explicit instructions, but the attack surface is real. A compromised Manus account could, in theory, give someone remote access to everything on your machine.

I'm not saying don't use it. I'm saying go in with your eyes open. For non-sensitive tasks (summarizing a public newsletter, analyzing a blog post), the risk is low. For forwarding client contracts or financial documents, think carefully about what you're comfortable sending to a third party.

Where this fits for developers building agent workflows#

If you're a developer building an AI agent that needs to work with email, it's worth understanding what category Manus falls into. Mail Manus is a consumer-facing feature. It's designed for end users who want to trigger tasks by forwarding emails manually. There's no API, no SDK, no way to programmatically integrate it into an agent pipeline.

The My Computer approach is creative but brittle for production use. Screen-based automation works for demos and personal productivity. It breaks down when you need reliability, speed, or scale. An agent clicking through Gmail's UI to send an email takes 30-60 seconds and fails if Gmail changes a button label. An API call takes milliseconds and fails predictably with error codes you can handle.

For developers who need their agents to actually own email addresses, send and receive programmatically, and handle incoming messages in real time, the architecture looks different. You want API-native email infrastructure where the agent provisions its own inbox and operates through code, not through screen clicks. LobsterMail is one option in this space, built specifically for agents to self-provision inboxes and handle email through an SDK rather than forwarding or GUI automation.

Should you use Manus for email tasks?#

For personal productivity, Mail Manus is genuinely useful. Forwarding a long thread and getting a summary back in two minutes saves real time. The My Computer feature is impressive as a demo and practical for occasional file-and-email tasks when you're away from your desk.

For anything involving sensitive data, high volume, or production reliability, you'll want purpose-built infrastructure rather than forwarding-based or screen-based approaches. The right tool depends on whether you're a user who wants AI help with email, or a builder who needs email as a capability inside an autonomous agent.