MCP (Model Context Protocol)
Also: MCP server, MCP connector, MCP integration. An MCP server is the part of a piece of software that offers its tools through this standard.
MCP (Model Context Protocol) is an open standard that lets an AI assistant use the tools and data of other software, for example to build a registration page in an event platform.
What MCP is not
| Term | What it is | How MCP differs |
|---|---|---|
| Language model | the AI itself, such as Claude or GPT | MCP doesn’t think; it connects the model to tools |
| API | an interface for programs that a developer calls | with MCP the assistant calls the tools itself, in the conversation |
| Plug-in | an extension for exactly one vendor | an MCP server works with any assistant that speaks the standard |
MCP and AI agents
An AI agent is an assistant that completes a task in several steps on its own: it plans, calls tools, checks the result and keeps going. MCP is where those tools come from. The agent is the MCP client; the software offering the tools is the MCP server. One agent can connect to several servers at once, say your calendar and your event platform.
| Term | What it is | Example |
|---|---|---|
| MCP client | the assistant or agent that uses tools | Claude, ChatGPT, a Copilot Studio agent |
| MCP server | the part of a piece of software that offers tools | an event platform’s server |
| Tool | an action with a name, description, inputs and outputs | “create page”, “build form” |
| Resource | data the client can read | a file, an API response |
How an MCP server exposes tools
The server describes each tool itself: its name, what it does in plain words, the inputs it needs and what it returns. The client reads that list when it connects and picks the tool that fits the request in the conversation. When the vendor changes its tools, the client sees the new list without anyone writing code.
A remote MCP server is reachable at a web address; the standard calls the transport “Streamable HTTP”. Access is usually handled by OAuth: the user signs in to the software, and the assistant then works with that user’s permissions, not with a master key of its own.
Microsoft Copilot Studio as an MCP client
Copilot Studio is Microsoft’s tool for building custom agents that can then run in places like Teams and Microsoft 365 Copilot. According to Microsoft’s documentation, you can add an existing MCP server to an agent as a tool, and the agent then uses that server’s tools and resources.
- Transport: Copilot Studio supports Streamable HTTP; the older SSE transport is no longer supported.
- Sign-in: none, an API key or OAuth 2.0; OAuth with “Dynamic discovery” is the simplest when the server supports dynamic client registration.
- Requirements: generative orchestration has to be turned on for the agent, and Power Platform data policies apply to MCP servers too.
MCP in event management
In an event platform, MCP means your assistant works where the event, the guest fields and the pages live, instead of handing you text to copy. It builds a registration page, creates the form from the guest fields, reviews the page on mobile and undoes a step when you ask. It does this with the permissions of the user who connected it, and every action is logged.
Three typical examples, depending on which tools a server offers:
- Build a registration page: the assistant reads the brief, creates the page and builds the form from the guest fields.
- Check a page before the invite goes out: the assistant renders the page on mobile and desktop and flags contrast problems or missing required fields.
- Two systems in one conversation: an agent connected to both your calendar and your event platform adds the event date to the invitation page.
MCP in AirLST
The AirLST MCP server
AirLST is MCP-ready. Airstage, the builder for landing pages and registration forms, is connected; other modules follow.
- Connect with one address, sign in with OAuth
- Assistants such as Claude, ChatGPT or GitHub Copilot; Microsoft Copilot Studio as a custom setup
- The user’s permissions, every action logged