MCP Server
An MCP server is a process that exposes tools, resources, and prompts to an MCP client over stdio, HTTP, or SSE following the Model Context Protocol spec.
In plain English
An MCP server is a small program whose only job is to expose some capability to AI apps. It advertises three kinds of things: tools the model can call (functions with typed inputs and outputs), resources the model can read (files, records, URLs), and prompts that package common workflows. An AI app connects, sees what is on offer, and calls what it needs.
Servers can run locally over stdio (great for filesystem or developer tools) or remotely over HTTP/SSE (great for SaaS integrations shared across users). They are usually small — a few hundred lines — because most of the heavy lifting is in whatever service they wrap. The community catalog includes servers for GitHub, Notion, Slack, Postgres, Google Drive, Figma, and dozens more.
Why it matters for Black Box
Black Box can point its specialists at any MCP server the owner has configured. An owner brings their own integrations — a Salesforce MCP server, an internal wiki MCP server — and the product consumes them without custom code. The protocol keeps the integration surface open.
Examples
- A local filesystem server exposing read/write/list tools over stdio.
- A remote Stripe server exposing customers, charges, and refunds as resources.
- A company-internal server exposing the HR system as a set of typed tools.