MCP Server

An MCP server is the half of a Model Context Protocol connection that a product runs. It publishes a catalog of tools, each with a title, a description and a typed input, and the connected assistant decides which ones to call.

Servers come in two shapes:

• Local: a process on your own machine, launched by the client, usually reaching local files or a local database
• Remote: a hosted endpoint the client reaches over HTTPS, authenticated with OAuth 2.1

Remote servers are the ones that matter for software-as-a-service products, because the data lives in the product rather than on the user's laptop. A well-built one separates reads from writes so a user can grant read-only access, marks tools that change data, shows a consent screen naming the client that is asking, and enforces the same limits the product's own interface enforces. Content the server returns from third parties, such as the text an AI engine produced about a brand, should be labelled as untrusted so the assistant treats it as evidence rather than instructions.

Prominara's server is remote and read-plus-safe-writes: it can list sites, read visibility, sources, missions and measurement verdicts, create tracked prompts, run a check, log a content change and set a mission status. Nothing deletes data or touches billing. See [the Prominara MCP server](/mcp) for the address and the full tool list.