An MCP server is the external system that supplies MCP tools to an AI application. An MCP client sends tool-call requests to server-side tool definitions for execution; one server can provide multiple tools. section 5.1.3
The 2026-07-28 MCP revision, commonly called MCP 2.0, removes protocol sessions in favor of self-contained requests. The mcp Python library 2.0 still represents connections through client sessions. section 5.1.3
Figure 5.4 The MCP architecture integrated into LLM agents and our framework

Shows how MCP servers, clients, and tools fit into an LLM agent framework. section 5.1.3
Figure 5.5 An LLM agent with tools coming from the two different types of MCP servers: stdio and streamable HTTP.

Compares the local stdio and remote streamable-HTTP server setups. section 5.1.3

















