Model Context Protocol (MCP)

The Model Context Protocol (MCP)⁠ is an open standard that lets large language model (LLM) apps, like Claude, ChatGPT, or Cursor, securely access user data from external services with the user's permission. Instead of asking users to sign in separately, MCP allows these AI apps to request permission to access specific data, such as emails or private Github repositories, directly through the app being used.

There are two parties involved in MCP - "client" and "server". In web development, the terms "client" and "server" often refer to the frontend (browser) and backend (web server). However, in this context, these terms have different meanings:

• The "client" is the LLM application that wants to access another service on a user's behalf. For example, Claude would be the "client" if it wants to get access to Gmail.
• The "server" is the system that hosts the protected resources the client wants to access. In this example, this would be Gmail. This is sometimes referred to as the "resource server" or "MCP server".

MCP servers often need to access user data from various sources on behalf of AI apps. This requires robust OAuth flows with proper consent management, token verification, and security controls, which Clerk supports. The combination of dynamic client registration (for registering MCP servers programmatically), the consent screen (for secure user authorization), and comprehensive SDK support makes Clerk an ideal authorization server for MCP implementations.

For step-by-step guides on working with MCP using Clerk:

• Build an MCP server
• Connect an MCP client