MCP Server
An MCP server is a program that exposes tools, resources, and prompts to AI applications through the Model Context Protocol. When a client connects, the two sides run an initialization handshake to agree on a protocol version and capabilities, after which the client can list what the server offers and call it over JSON-RPC 2.0. A server differs from the MCP client, which lives inside the host application and opens the connection, and from a plain REST API, which a server often wraps but which has no standard way to describe itself to a model. In agentic systems, MCP servers are the usual way to give an agent access to files, databases, SaaS products, or internal services without writing a new connector for each model or framework.
Servers run locally as a subprocess over standard input and output, or remotely over Streamable HTTP, where the specification builds authorization on OAuth 2.1. The same server works with any MCP-compatible client, so one server for a ticketing system can be reused across chat assistants, IDE agents, and background workflows. Because a server can act on real systems, the host should treat its tool descriptions and returned content as untrusted input, and write operations usually sit behind scoped credentials or an approval step.
Related terms
Related services: Agentic AI consulting.
Ready to put agentic AI to work?
Book a free 45-minute consultation. We'll map one real process worth automating with production-grade AI.