What is the Model Context Protocol (MCP)?
The Model Context Protocol, shortened to MCP, is an open standard that lets an AI client such as Claude use tools from other systems. The system runs an MCP server that offers those tools, the client connects with a token and calls them. In PR that means searching journalists, building media lists and scheduling sends from one conversation.
Written by Timon Hendriks · Last updated on 20 August 2026
How it works in practice
An MCP connection has two sides. The server belongs to the system holding your data and describes which tools exist, what input they expect and what they return. The client is your AI environment: Claude Desktop, Claude Code, Cursor or Zed. You paste one server URL and an access token into that client and the tools appear. The token carries permissions, so you decide whether a client may only read or may write as well.
For communications work this mostly changes the order of things. Instead of walking through screens yourself, you describe the outcome you want: a media list for a product launch, a draft based on earlier announcements, or an overview of who opened the last mailing. The AI client picks the tools, runs them and shows what is ready. You review it and approve, above all before anything is sent.
Example
A communications agency with fourteen clients connects its PR system to Claude over MCP. On Monday morning the account manager asks for an overview of every campaign scheduled this week, flagging the clients without a media list. Claude calls the search and overview tools in sequence and returns the list. The account manager then has media lists proposed for two clients, edits them by hand and queues the send for Thursday at 07:00.
Common mistake
Assuming an MCP connection means the AI starts sending on its own. A sound setup separates preparing from sending: the release and the recipients are queued, a human confirms. Agree on that line before handing out a token with write access.
Frequently asked questions
What does MCP stand for?
MCP stands for Model Context Protocol, an open standard Anthropic published at the end of 2024. It describes how an AI client may request and use tools and data from other systems, whichever vendor supplies them.
Do you need a developer for an MCP connection?
Usually not. If the system already offers an MCP server, you connect with an access token and a single line of configuration in your AI client. You only need a developer when you want to build a server for a system that does not have one yet.