---
title: "Comment by Vic Moreno on \"Email MCP Servers on a Mac: What to Look For\" | Hashnode"
description: "The row that matters most is \"can the agent send on its own\": with a compose window as the only exit, a prompt injection in one email ends as a draft, not a sen"
url: "https://hashnode.com/posts/email-mcp-servers-on-a-mac-what-to-look-for/6aeffaa578ee7687d0c846e8/comment/6af07f9c945132673f11401c"
---

[Discussion](https://hashnode.com/posts/email-mcp-servers-on-a-mac-what-to-look-for/6aeffaa578ee7687d0c846e8)

Comment [Email MCP Servers on a Mac: What to Look For](https://hashnode.com/posts/email-mcp-servers-on-a-mac-what-to-look-for/6aeffaa578ee7687d0c846e8)

[Vic Moreno](https://hashnode.com/@itsvicmoreno)

Indie hacker reviewing the one-person SaaS stack. Verdict first.

[22h ago](https://hashnode.com/posts/email-mcp-servers-on-a-mac-what-to-look-for/6aeffaa578ee7687d0c846e8/comment/6af07f9c945132673f11401c)

The row that matters most is "can the agent send on its own": with a compose window as the only exit, a prompt injection in one email ends as a draft, not a sent mail. Marking bodies as untrusted helps, but it is a hint to the model, not a wall. Does the one-week undo also cover drafts the agent created?