AI 시대의 필수 지식 MCP 이 영상 하나로 끝내세요!

Summary

The video introduces the Model Context Protocol (MCP), demonstrates how to use an existing MCP server, shows how to build and deploy a simple one, and presents a sponsored platform for sharing MCP-connected AI with a team.

What MCP is and why it matters

A large language model generates text based on patterns learned during training. On its own, it may not know current or private information—such as today’s weather or a company’s internal policies—and may produce incorrect answers, or “hallucinations.”

Connecting AI to external tools and data lets it retrieve information or carry out actions instead of relying only on its training. Developers can connect AI to services through APIs, but building and maintaining each integration separately can be cumbersome.

MCP is a common protocol—a shared format or set of rules—for connecting AI clients to tools and data sources. The video compares it to USB-C: tools that follow the standard can connect more consistently to compatible AI clients. MCP is not the only way to connect AI to tools; its value is standardizing those connections. The video says Anthropic introduced MCP in November 2024.

Examples of MCP in use

• Blender: The presenter describes connecting Claude Desktop to Blender through an MCP server. The AI can then manipulate Blender to create 3D objects, even if the user does not know how to operate the software.
• Other integrations: The video mentions design tools such as Figma, note-taking tools such as Notion, and calendar services.
• Context7: This MCP server can retrieve development documentation, such as current Next.js documentation, so an AI can answer coding questions using that material rather than relying only on its built-in knowledge.

Connecting an existing MCP server to Claude Desktop

The video demonstrates this workflow with Context7:

1. Install and open the desktop version of Claude.
2. In Claude’s settings, open the Developer area and locate the MCP connection options.
3. Find an MCP server through a directory such as Smithery and select Context7.
4. Copy the configuration for the relevant connection method. The demonstration uses a Windows configuration.
5. Open Claude Desktop’s configuration file, paste in the MCP server settings, and save.
6. Restart Claude Desktop so it loads the server.
7. Confirm that Context7 appears among Claude’s available tools.
8. Ask a question that requires the tool—for example, how to create a Next.js app.
9. Review and approve the tool call when prompted. Claude retrieves the documentation and uses it to answer.

MCP servers can run locally or be hosted remotely.

Creating and connecting a simple MCP server

The video uses Python and FastMCP to demonstrate building a server:

1. Load the FastMCP library in a Python project.
2. Define a function for the desired task. The example is an addition function that accepts two numbers and returns their sum.
3. Register the function as an MCP tool using FastMCP’s tool mechanism.
4. Run the server. The example listens locally on port 8000.
5. Add the server to an MCP client such as Visual Studio Code. The demonstration opens the command palette with Ctrl+Shift+P, chooses the MCP server options, and supplies the server connection details.
6. Ask the AI to perform the calculation using the new tool. It identifies the two inputs, sends them to the server, and uses the returned result in its answer.

The presenter explains three connection methods:

• STDIO: The client communicates with a program running locally through standard input and output.
• Streamable HTTP: The client communicates with a server over HTTP, with streaming support. This is the method used in the example.
• SSE: An older, less commonly used option.

Making an MCP server available remotely

To let others use a server, deploy the Python application to a hosting service as you would a web application. Then replace the local server address in the client configuration with the deployed server’s address. Users can access the remote server from other computers without the developer’s local server running.

Team use through a managed platform

The video’s sponsored segment presents a platform for making MCP-connected AI easier to share with coworkers who may not want to configure MCP themselves. The transcript calls the platform “Porter AI”; the video metadata identifies the sponsor as Ale AI.

The demonstrated capabilities include:

• Choosing among different AI models and adding models through OpenRouter.
• Connecting MCP servers, including one for company user data and one for Notion-based company policies.
• Asking questions that require internal information—for example, about a policy or recent subscriber counts—and having the AI call the relevant MCP tool.
• Inviting teammates, setting permissions, and reviewing model usage statistics.
• Integrating the AI with Slack so employees can ask questions by tagging it in a channel instead of opening a separate AI interface.

These features are presented as ways to make company MCP tools accessible to nontechnical teammates. The video directs viewers to the platform’s documentation for further instructions.

Speakers and sources featured

• Presenter: JoCoding
• AI client/model shown or discussed: Claude / Claude Desktop
• Organizations and services mentioned: Anthropic, Smithery, Context7, FastMCP, Python, Visual Studio Code, Cursor, Blender, Figma, Notion, Slack, OpenRouter, and the sponsored platform identified in the metadata as Ale AI (rendered as “Porter AI” in the subtitles)
• Illustrative data/tool example: A hypothetical connection to a weather service or weather data source