Model Context Protocol
Also called MCP
An open standard that lets an AI agent use outside services, such as a mailbox or a calendar, through a fixed set of tools that the service defines.
An AI agent on its own knows only what you type into it. The Model Context Protocol gives it a standard way to reach other services. The service runs an MCP server that offers a list of tools, such as “search mail” or “create a task”. The agent calls those tools when your request needs them.
The service decides which tools exist, so it also decides the limits. An agent cannot do anything the server does not offer. You approve the connection and its permissions when you add the server to your agent, and you can remove it later.
For email, the limits matter more than the features, because any stranger can send text to your mailbox for the agent to read. In Kistrl, a connected agent can read what you allow and write drafts, notes and tasks, but it can never send mail or delete anything. See Agents over MCP and read How to let an AI agent work with your email safely.