MCP Server Learning

The project mainly learns and implements the relevant functions of the MCP protocol, including core concepts such as sampling and root definition, explores two transmission methods (stdio and sse), and attempts to build a Node-based server.

Artificial intelligence chatbots Developer tools #Protocol learning #Data transmission #Server development #SDK application .TypeScript

rating : 2 points

downloads : 13

update time : 2025-04-29

What is an MCP server?

The MCP server is an implementation of the Model Context Protocol, serving as a bridge between AI applications and large language models. It provides a structured way to manage context resources, predefined prompts, and execute tool operations.

How to use an MCP server?

You can connect in two ways: 1) Run the server directly through standard input/output (stdio). 2) Communicate over the network via Server-Sent Events (SSE).

Use cases

Suitable for applications that require structured AI interactions, such as intelligent assistants, automated workflows, and content generation systems. Particularly suitable for enterprise environments that require security boundaries and resource management.

Main features

Resource managementManage structured data such as PDFs and database records to provide context references for AI.

Predefined promptsStore and manage commonly used prompt templates to improve interaction efficiency.

Tool invocationAllow AI models to perform specific operations (such as file processing, API calls, etc.) through the server.

Sampling functionControl the diversity and randomness of AI output (a feature under development).

Security boundariesDefine the physical and logical boundaries of server operations to ensure security.

Advantages and limitations

Advantages

Built-in error handling mechanism for easier development

Flexible choice of transport protocols (stdio/SSE)

Comprehensive security boundary control

Clear API specification (JSON-RPC 2.0)

Limitations

Limited compatibility with some frameworks (such as fastify)

Additional network configuration required for SSE mode

Steeper learning curve

How to use

Select a transport method

Choose the stdio or SSE communication method according to your needs.

Configure the server

Set the resource path, tool functions, and security boundaries.

Define the specific operation functions that AI can call.

Start the service

Run the server and connect the client.

Usage examples

File processing assistantQuery and process files in a specified directory through AI.

Database query proxySafely query the database without exposing credentials.

Frequently Asked Questions

Why choose SSE instead of WebSocket?

How to ensure the security of tool invocation?

Can it be integrated with other frameworks (such as Fastify)?

Related resources

MCP official documentation

Protocol specifications and API references

GitHub example library

Code examples for various use cases

JSON-RPC 2.0 specification

The communication protocol standard used by MCP

🚀 Learning the MCP Protocol

This project focuses on learning the MCP protocol, including implementing related functions and understanding key concepts.

🚀 Quick Start

This README is about learning the MCP protocol, including tasks to be completed and key learning points.

✨ Features

Implement sampling method: https://modelcontextprotocol.io/docs/concepts/sampling
Implement root directories.

📚 Documentation

To - Do List

Implement the sampling method at https://modelcontextprotocol.io/docs/concepts/sampling.
Implement root directories.

Learning Points

There are two transmission methods:
- Via stdio. Here, you need to provide the path to a server binary for the client to use.
- Via sse. Here, you need to provide the network address of the server.
  - This transmission method seems more versatile to me. It allows the server to be deployed independently of the client.
I really wanted to use fastify to build the server, but I couldn't get it to work with the existing SDK.
- This is a great opportunity to learn how to write such a server using only Node.js!
Core primitives related to MCP include:
- Resources – Think of these as data (e.g., PDFs, database records, etc.) that can be used to shape the correct response based on user queries.
- Prompts – Consider these as the prompt statements defined below the text box sometimes seen in the "prompt - to - x" process.
- Tools – These allow the LLM to perform actions on behalf of the user. The server is responsible for invoking tools and using the results to respond to the client. The client's task is to pass these results to the LLM.
- Sampling – I'm not entirely sure about these yet.
- Root directories – They define the "boundaries" of the server. For example, if your server exposes a tool that interacts with the file system, you can "restrict" this tool to a specific directory by including it in the "root directories".
- Transmission methods – They have been mentioned above. There are two transmission methods, STDIO and SSE. As for the message format, it is based on the JSON - RPC 2.0 protocol.
I really like the built - in error handling in the SDK.
- If I throw an error in the server tool callback function, the client SDK will handle the error and respond accordingly.