What Is an MCP Server? Definition, Examples, and Hosting

An MCP server is a program that exposes tools, resources (data), and prompt templates to an AI application over the Model Context Protocol (MCP), with tool discovery through tools/list and tool invocation through tools/call; resources and prompts have their own methods. MCPifex hosts real MCP servers behind one gateway endpoint instead of you installing and running the package yourself.

Server, client, and host — three different things

MCP's architecture has three participants, and "MCP server" names only one of them. The host is the application a person uses — Claude Code, Claude Desktop, ChatGPT. The client is the connector inside that host holding one dedicated connection to one server. The server is the separate program on the other end that actually does the work — querying a database, reading files, calling an API — whether it runs as a local subprocess over stdio or as a remote process reachable over Streamable HTTP.

Why it matters

Keeping the roles straight matters for two reasons. First, it's the server's exposed tools, resources, and prompts that a client actually calls — a broken tool call or an over-broad permission belongs to one server, not to the whole host application. Second, it's why a client built to the MCP spec needs no code changes to talk to a hosted server: MCPifex runs the server for you over the standard Streamable HTTP transport at https://mcpifex.com/mcp, so you create an instance in the portal, choose which tools are enabled, and point an existing MCP client at it instead of installing and running the package yourself — starting on the free plan.

Related terms

The MCP explainer covers the relationship between these terms.

See MCP client for the connector on the other end of that connection, and Model Context Protocol for the umbrella spec that defines all three roles.

As of September 2026, the MCPifex configuration described here uses instance API keys and the hosted gateway’s documented tool controls.