TheModelDesk

The Model Context Protocol (MCP) serves as a standardized interface for exposing tools, resources, and prompts to AI agents. Much like REST did for web APIs, MCP allows developers to create tools once as servers that any compatible client can reuse, whether hosted locally via stdio or remotely via HTTP.

This tutorial demonstrates how to implement an MCP architecture using FastMCP, a Python library that simplifies server creation by using decorators similar to FastAPI. To bridge these servers with an AI agent, the project utilizes langchain-mcp-adapters, which allows LangChain v1 to integrate tools from multiple sources seamlessly.

The technical stack includes Ollama for running the Qwen model (specifically qwen3.5:4b), which provides native tool-calling capabilities. The example setup consists of a local FastMCP server providing basic utilities like word counting and time tracking, and a remote server from DeepWiki for accessing GitHub repository data.

By separating the tools into standalone servers, the agent becomes a coordinating script that merges tools from both local and remote sources. This approach ensures that the core LLM remains local while selectively interacting with external services only when necessary.