What is the Model Context Protocol (MCP)?

Before MCP, connecting an assistant to a product meant a bespoke integration per assistant. MCP replaces that with one interface: a product exposes an MCP server, and any compatible client can use it.

For the user this shows up as being able to do real work from inside the assistant they already have open. Rather than switching to another application, you ask, and the work happens in the underlying product.

Access is scoped by key, which matters when the underlying product holds several clients or brands. A key limited to one brand can only act within it, and revoking the key ends the access.

It is a plumbing standard rather than a feature, which is why the interesting question is never whether a product supports MCP but what it lets you actually do once connected.

What that looks like in practice is a request that crosses several steps without moving between windows. Asking which posts beat your average last quarter, and then asking for three drafts in the voice those posts established, becomes one conversation rather than a report, a copy, and a paste. The assistant is not guessing at your numbers either; it is calling the product and getting the real ones back.

None of this removes the approval step. An assistant acting over MCP can do only what the key it was given allows, and a product that requires a person to approve a post still requires it when the request arrived through a chat window.