MCP uses a client-server architecture where:
MCP Servers expose tools or data (e.g., a get_weather tool or a file system). MCP Clients (AI applications or Agents) connect to these servers, discover available tools, and invoke them via standardized calls (built on JSON-RPC). The LLM can then decide when and how to use these tools based on the task.

MCP Servers expose tools or data (e.g., a get_weather tool or a file system).
MCP Clients (AI applications or Agents) connect to these servers, discover available tools, and invoke them via standardized calls (built on JSON-RPC). The LLM can then decide when and how to use these tools based on the task.