You’ve probably noticed AI assistants getting more useful lately — not just answering questions, but actually doing things: reading a document, checking a calendar, sending a message. The thing that makes that possible has a name, and it’s worth understanding, because it changes what your email can be.
It’s called MCP — the Model Context Protocol.
The plain-English version
An AI model on its own is a closed box. It’s clever, but it can’t see your inbox, your files, or your calendar unless something feeds them in. For a long time, connecting an assistant to a real service meant building a one-off, bespoke integration for that exact assistant and that exact service — slow, brittle, and different every time.
MCP is an open standard that fixes that. Think of it as a common plug — like USB for AI. A service (say, an email provider) exposes a small, well-defined set of capabilities over MCP: “list recent messages”, “search”, “read this one”, “send a reply”. Any MCP-capable assistant can then use those capabilities without anyone writing custom glue for that specific pairing.
One standard, many assistants, many services — instead of a tangle of private integrations.
Why an open standard matters
The “open” part is the whole point, and it’s easy to undervalue:
- You’re not locked to one assistant. Because MCP is a standard, the same mailbox works with whatever MCP client you choose — and can keep working with the next one, without your provider rebuilding anything.
- Capabilities are explicit. MCP makes the assistant’s abilities a defined list — these actions, nothing hidden — rather than vague, total access. That’s easier to reason about and to secure.
- It’s inspectable. An open protocol can be audited and understood, instead of being a black-box connection you have to take on faith.
Bespoke integrations quietly lock you in. An open protocol does the opposite.
What “email for AI agents” actually means
Put those two ideas together — email and MCP — and you get something new: a mailbox an assistant or an automated agent can genuinely operate. Not a screen-scraper pretending to be you, and not a throwaway inbox with no real deliverability, but a proper mailbox that exposes real actions over the standard.
In practice that means an assistant can triage your inbox each morning, an agent can watch for a specific kind of message and act on it, or a workflow can send a reply in-thread — all through the same open interface, with the capabilities defined up front.
The important part is how that access is granted. Reading your mail is powerful, so it has to be done safely — scoped, revocable, and with hard limits on what an agent can do in your name. (We wrote about exactly that in “Giving an AI assistant access to your inbox, safely”.)
Where Lettio fits
Lettio is email built for this from the ground up. It ships a native MCP server — hosted with OAuth at mcp.lettio.eu, or run locally as @lettio/mcp — so any MCP-capable assistant can read, search, reply and organise your inbox over the open standard. Sending is pinned to your own address, so an agent can act for you but never as someone else.
And because it’s a real, EU-hosted mailbox behind that interface — not a bare relay — your mail stays private, under EU jurisdiction, and yours to export, whether a human or an agent is reading it.
Curious how it connects? See the developer docs, or read what Lettio protects.