An MCP server is a program that exposes tools, resources, and prompts to AI applications through the Model Context Protocol. The AI client connects to it and calls those capabilities on the user's behalf.
How an MCP server works
An MCP server sits on the other side of a connection from an MCP client such as Claude Desktop, Cursor, or VS Code. When the client starts a session, it asks the server what it offers. The server answers with three kinds of capability:
- Tools: functions the model can call, such as
draft_postorlist_posts. - Resources: read-only data the client can load into context.
- Prompts: reusable templates the user can pick.
Messages use JSON-RPC 2.0. The connection runs over one of two standard transports: stdio for a local process, or streamable HTTP for a remote service.
An example
You ask your assistant to "draft three posts about Friday's release." The model sees a draft_post tool in the server's tool list, fills in the arguments, and calls it. The server does the work and returns a result the model can read back to you.
Why it matters for social publishing
An MCP server is where you decide what an agent may do. A server that only exposes drafting tools cannot publish, however the model is prompted. SocialHandoff runs as an MCP server over both stdio (npx socialhandoff-mcp) and streamable HTTP. Its tools let an agent draft, schedule, and read insights, while approval stays with a person holding an owner key.
For the longer walkthrough, read what is an MCP server, or set one up with the MCP config generator.