MCP (Model Context Protocol)

MCP (Model Context Protocol) is an open protocol by Anthropic that standardizes how LLM-powered applications communicate with external data sources and tools. It defines three main “primitives”:

• Resources (like GET endpoints, e.g., resource://food/pizza)
• Tools (functions callable by an LLM, e.g., search_docs(query="..."))
• Prompts (reusable conversation templates)

This repo shows multiple examples how to integrate servers and clients into AI Apps.

Example Project Files:

• client_resource_langgraph.py: Extends MultiServerMCPClient into MultiServerMCPClientWithResources, allowing listing and reading MCP resources, demonstrated by a simple example.

• client_resource.py: Connects to an MCP server via SSE, lists available resources and prompts, reads a specific resource, and demonstrates retrieving a configured prompt.

• host_and_client.py: Uses the MultiServerMCPClient with a LangGraph agent. Loads tools from the MCP server and enables an LLM model (GPT-4o-mini) to autonomously invoke those tools (e.g., "What food does Bella Vista offer?").

• server.py: Sets up an MCP server accessible via SSE, defining tools (add, multiply, search_docs), resources (resource://food/{item}), and a prompt (friendly_greeting).