Atomic Mail Blog: email infrastructure for AI agents

An MCP server gives an AI app tools and data through one standard protocol. How it works, local vs remote, real examples, a working Python server, and security risks.

MCP vs API explained: how the Model Context Protocol relates to REST APIs, token costs, auth, CLI as a third option, and one task built both ways in code.

Gmail was built for people, not agents. OAuth tokens that expire, restricted scopes, sending caps and setup overhead explain why, and what an agent-native inbox looks like

Send email in Python with smtplib and EmailMessage: plain text, HTML, attachments, Gmail app passwords, port 465 vs 587, and when to switch to an email API

IMAP needs several round trips to list and read five emails. JMAP does it in one. What the protocol changes, and who already uses it.

Don't hand OpenClaw your own Gmail. Here's how an OpenClaw agent registers an inbox of its own on Atomic Mail, and the cron job that actually makes it read.

Six value types, double quotes only, no trailing commas. What JSON is, how to read and write it, and why it beat XML.

An agent can't tell your instruction from one hidden in an email it reads. What that breaks, and how to build so it doesn't matter.

Client, request, server, response. What an API actually is, the four main styles, and why AI agents run on them.