What this screen is for
MCP server documents how an AI assistant can operate MailGuardHQ on your behalf. The Model Context Protocol is the standard way an assistant calls a service; our server exposes the same REST API as a set of tools an assistant can pick from.
The page exists so you can decide whether it is worth waiting for. The server is not released. There is no address to point a client at, no published repository and no registry listing - running it today means running it yourself, from source.
What each part of the page does
- What it is
- Why the server holds no logic of its own. Every permission decision stays in the API, so an assistant can never do more than you could with the same token.
- Connecting
- The three steps - a token, a running server, a client pointed at it - with the configuration a client needs.
- Authentication
- What the server refuses before it forwards anything, and why each refusal exists.
- Scopes
- The two things an MCP token can never do, whatever scopes it names: change your DNS, and grant somebody administrative access.
- The tools
- The full tool list, generated from the API contract rather than written by hand.
- What is not built yet
- Five open gaps, stated plainly. One of them means the domain tools do not work with an MCP token today.
What to do next
- Read "What is not built yet" first. It is the part that decides whether this is useful to you today.
- If you want to try it now, mint a token for the mcp audience from your organisation and run the server yourself against the live API.
- If you would rather wait, tell us what you would connect it to - what ships first is driven by what people ask for.
What this screen does not do
It describes a server we do not host for you
There is no public endpoint. Nothing on this page is a link you can connect a client to, and no part of the product depends on the MCP server working.
The free domain check, the web app and the REST API are unaffected by any of it - the MCP server is another way in, not the way in.