Understanding MCP From Scratch

The Model Context Protocol (MCP) is an open standardized protocol that connects AI models with various data sources and tools. It functions like a "USB-C port for AI applications," allowing LLMs to seamlessly interact with local files, databases, APIs, and custom tools. But, it can be challenging to understand. Here, we breakdown how MCP works, build an MCP server for LangGraph docs from scratch, and connect it to Claude, Cursor, and Windsurf. Video notes: https://mirror-feeling-d80.notion.site/MCP-From-Scratch-1b9808527b178040b5baf83a991ed3b2?pvs=4 Chapters: 00:00 - Introduction to MCP 00:21 - Demo: MCP Tool in Cursor 00:40 - Demo: MCP Tool in Windsurf 00:54 - Demo: MCP Tool in Claude Desktop 01:15 - Motivation for MCP 01:38 - Building a Tool from Scratch 02:00 - Loading LangGraph Docs and Creating a Vector Store 02:44 - Testing the Basic Vector Store Query 03:00 - Creating a LangChain Tool 03:22 - Binding Tools to LLMs 04:00 - Bridge to MCP: Connecting Tools to AI Applications 04:24 - MCP as a Client-Server Protocol 05:00 - How MCP Works with Host Applications 05:37 - Server Initialization by Host Applications 06:05 - Defining an MCP Server 06:41 - Adding Resources to MCP 07:05 - Running the MCP Inspector 07:27 - Testing Tools in the Inspector 08:06 - Configuring MCP Servers for Different Hosts 08:48 - Demo: MCP Tool in Cursor (Revisited) 09:10 - Demo: MCP Server in Windsurf (Revisited) 09:32 - Demo: MCP in Claude Desktop (Revisited) 09:56 - Using MCP Resources in Claude Desktop 10:19 - Recap: Augmenting LLMs with Context and Tools 10:57 - Conclusion and Final Thoughts