MCP server

MCP, for Model Context Protocol, is an open protocol letting a conversational assistant call external tools. A software vendor exposes its functions as tools described in a standard way; the assistant discovers them, understands their parameters, and calls them when the conversation requires it.

What it changes in practice

Before, automating a task meant writing code against an API. With an MCP server connected, someone who does not program can ask in plain language to voice the last three sports articles, or how many credits are left this month, and the assistant turns that into calls. The server acts with the rights of YOUR key: it can do nothing you could not do yourself.

For a newsroom, the point is not the novelty. It is that driving a daily production stops requiring either a developer or a trip into an interface. People already work in a conversation; the tool comes to them.

The question that matters, and that nobody asks

Parity. An MCP server exposes a number of tools that is almost never equal to the number of API operations. 'MCP compatible' is therefore an empty claim until you know what is covered.

We were caught out by our own copy on this point, which is instructive: four pages of this site claimed our server gave an assistant everything a key allows. Ten tools for twenty operations at the time: half had no equivalent, including targeted retouching, one of our five distinctive features. A customer who chose us to drive production from an assistant would have found out while trying to fix a typo. The coverage table is now COMPUTED from the code, which stops it from quietly becoming false again.

Ask your vendors the same thing: not whether they do MCP, but which operations the assistant can perform, and which remain reserved for code.