What MCP is
Without a standard, every AI model needs its own integration with every system. MCP replaces that with one shared interface. A system is wrapped once as an MCP server, and any assistant that speaks MCP can use it.
MCP is supported by the main AI platforms, including Microsoft’s, so a server you build today is not tied to one model or one vendor.
How it works
There are three parts. The host is the application a person uses, such as an assistant in Microsoft Teams. Inside it, an MCP client connects to one or more MCP servers. Each server exposes a system to the assistant.
A server can offer tools, which are actions such as creating a ticket or drafting a reply. It can offer resources, which are things to read such as a document or a record. It can also offer prompts, which are templates for common tasks. Each is controlled differently. The assistant decides when to call a tool, the application decides which resources to give it, and a person chooses a prompt. The server does the work, with its own credentials and its own rules.
What it lets you do
An assistant can look up a case in your CRM, read the relevant documents in SharePoint, draft a reply in Outlook and file a note, each step through a server you control. The same servers can be used by an agent that runs without a person typing, with a person approving the steps that matter.
Where the risks are
A tool that can act can act wrongly. The main risks are permissions that are wider than the task needs, servers from sources you have not checked, and prompt injection, where text inside an email, web page or document tries to instruct the assistant to do something else.
Controls to set before you connect anything
Give each server only the access its task needs, and keep reading and changing things as separate tools. Require a person to approve any action that changes data or sends something outside the organisation. Log every call, with who asked for it and what it did.
Run the servers that touch your data in your own tenancy, check any third party server before you use it, and test the assistant with content written to mislead it before real users see it.
Where to start
Start with one read only server on one system and one group of users. Once the answers are useful and the logs look right, add actions one at a time, each with an approval step.
How Sendient approaches this
We connect email, calendars, Teams, documents and business systems to AI through MCP and APIs, on Microsoft Azure, with permissions, approvals and an audit trail in place from the start. See AI integration and MCP, AI agents and automation and Azure AI.
