The Model Context Protocol (MCP) is an open standard for connecting AI applications to external tools and data. It defines how a client discovers and calls capabilities that a server exposes.
What MCP defines
Anthropic published the Model Context Protocol as an open specification in November 2024. Before it, every AI app wired up integrations its own way. MCP gives them one shape: a host application runs one or more clients, and each client holds a session with one server.
The protocol covers:
- Capability discovery, so a client can list a server's tools, resources, and prompts.
- Calls and results, framed as JSON-RPC 2.0 messages.
- Transports: stdio for local processes and streamable HTTP for remote ones.
- Authorization for remote servers, based on OAuth.
An example
The same server can be added to Claude Desktop, Cursor, and Gemini CLI without changes. Each client reads the server's tool list and lets its model call those tools.
Why it matters for AI-agent publishing
With MCP, you write the integration once and every compatible assistant can use it. It also puts the permission boundary in the server rather than in the prompt, which is where a social account needs it.
SocialHandoff implements MCP with tools for planning, drafting, scheduling, previewing, and reading insights, plus MCP Apps UI cards for a calendar, an approval view, and analytics. The connect an agent with MCP guide covers setup, and the MCP server entry explains the server side in more detail.