What is an MCP server?
A program that exposes tools, resources or prompts to AI applications over the Model Context Protocol.
A local server runs on your machine as a subprocess of the AI app (stdio transport) and can reach local files or programs. A remote server runs on the internet behind one HTTPS URL (Streamable HTTP transport); you connect it by pasting that URL into your client, with nothing to install.
A server declares what it supports during initialization. Tools are the most widely used primitive: functions with a name, a description and a JSON Schema for their inputs that the model can choose to call.
Sources
Related terms
- MCP client and hostThe host is the AI application (an IDE, a desktop chat app, a CLI agent); inside it, one MCP client holds the connection to each MCP server.
- Streamable HTTP transportThe MCP transport for remote servers: one HTTPS endpoint (such as https://example.com/mcp) that receives JSON-RPC messages as POST requests and may stream replies as server-sent events.
- Tool call (tools/list, tools/call)How an AI model uses an MCP tool: the client lists the server's tools with tools/list, the model picks one, and the client runs it with tools/call and returns the result to the model.
- MCP RegistryThe official, open catalogue of MCP servers at registry.modelcontextprotocol.io, where publishers list a server.json describing how to install or connect each server.