Recently, MCP has really gained a lot of popularity. It's like giving wings to AI, enabling it to call various external tools to serve you. However, just like getting a new toy with a rather complicated instruction manual, many people find it difficult to configure.

Don't worry. This article will lead beginners with zero foundation into the world of MCP in the simplest way. After reading, you'll surely learn how to use MCP services from scratch!

Without further ado, let's get started!

1. What is MCP and what are its functions?

1. What is MCP

MCP stands for Model Context Protocol, a brand - new technology open - sourced and released by Anthropic (the parent company of Claude) in November 2024. It is a standardized protocol that allows large AI models to interact with various external services and resources.

Its core function is to unify the data interaction format standard between large - scale language models (LLMs) and external tools. The MCP protocol is like a universal translator or a unified interface standard. It specifies the way various applications transmit data to AI. As long as an AI model is compatible with the MCP protocol, it can interact with all MCP - supported applications without any obstacles.

To put it simply:

You can think of MCP as the USB - C interface in the AI field, acting as a universal expansion dock. Various tool applications such as file managers, calendars, chat software, and music - playing software can be connected to large models through MCP.

If you still can't understand after reading this, here's a more appropriate example that will help you get it:

MCP is equivalent to equipping large AI models (the "brain") with "hands" and "feet", making it easier for large models to call and use external tools, services, or data. As a result, AI can truly help people with their work.

2. What are the functions of MCP?

After all that talk, to sum up the functions of MCP in one sentence: It enables AI to smoothly call various external services, greatly expanding the ability boundary of AI.

This is like equipping AI with a variety of handy toolboxes, transforming it from a "smart brain that can only chat and answer questions" into a super assistant that can actively solve practical problems for you and provide customized services.

MCP adopts a client - server distributed architecture, dividing the communication between LLMs and resources into three main parts: clients, servers, and resources.

• **MCP Host (Host Application):** Hosts refer to the applications that initiate and connect to LLMs, such as applications like Cursor, Claude, Desktop, and Cline.
• **MCP Client (Client):** The client is used to maintain a 1:1 connection with the Server within the Host application. Multiple MCP clients can run in one host application, thus connecting to multiple different servers simultaneously.
• **MCP Server (Server):** An independently running lightweight program that provides context, tools, and prompts for clients through a standardized protocol, and is the core of MCP services.

2. Preparations before configuring and using MCP services

After understanding what MCP is in general, next we'll start with the preparations needed before configuring and using MCP services.

1. Currently, there are mainly two modes for configuring MCP services:

• Stdio Mode: This mode is mainly used to connect software or files on your local computer. For example, if you want AI to control software like Blender that doesn't have an online service, you have to use Stdio. Its configuration is relatively more complex.
• SSE Mode: This mode is used to connect online services with their own APIs. For example, accessing your Google Mail, Google Calendar, etc. The configuration of SSE is extremely simple, basically just one link does the trick.

2. Preparations before configuring MCP services

The problem is that configuring MCP, especially in Stdio mode, has a relatively high threshold for ordinary people, and some preparations are needed. However, don't worry. Just follow me step by step, and you can handle it!

To configure MCP in Stdio mode, we need to use some command - line tools. We mainly need the uvx and npx commands. The following configuration operations need to be carried out on your computer:

1) Install uv (the uvx command is included):

Windows System:

• Press the "Win" key on your keyboard.
• Type "PowerShell" in the popped - up search box.

• Find the "Windows PowerShell" or "PowerShell" application, right - click on it, and select "Run as administrator". This step is very important!

• In the opened blue or black window, paste the following command and then press Enter to run it:

powershell -ExecutionPolicy ByPass -c "irm <https://astral.sh/uv/install.ps1> | iex"

• After running it, remember to restart your computer.
• Mac System:

2) Install npx (you need to install Node.js first):

The npx command is included in Node.js. Installing Node.js is very simple. The operations are as follows.

• Visit the official Node.js website: **https://nodejs.org/**.
• Click the download button (usually the current stable version is recommended) to download the installation package.

• Just like installing ordinary software, double - click the downloaded installation package and follow the prompts to click **"Next"** to complete the normal installation.
• After the installation is complete, your computer will have both Node.js and npm (Node.js's package manager, and npx is included in npm) [This comes from external common knowledge. Although not in the original material, it is standard to install npm and npx when installing Node.js].

Congratulations! After completing these two steps, the most basic computer environment preparations are done, and the subsequent configuration will be much easier.

Special Reminder: The source code mentions that if you're a Windows user without programming experience, it's very normal to encounter errors due to environment configuration or network issues. It may not be your fault. If you really can't figure it out, don't force it. Mac usually has fewer problems in this regard.

3. Where to obtain MCP services?

Well - begun is half - done. After setting up the "shovel" and "basket" (the basic environment deployment), now it's time to find the "ingredients" we need - MCP services!

As MCP becomes more and more popular, there have emerged many MCP aggregation websites. These websites are like the "app stores" for MCP services, where you can find the MCP services you need.

1. Aggregation sites providing MCP services:

1) Domestic Sites:

• AIbase: https://mcp.aibase.cn/explore

• Magic Tower MCP Square:https://modelscope.cn/mcp
• Baidu MCP Square: https://sai.baidu.com/mcp
• Nano **MCP Universal Toolbox:** Download address https://bot.n.cn/download?src=AIBotCode

2) Foreign Sites:

• MCP.so: https://mcp.so/
• Smithery.ai: https://smithery.ai/
• glama.ai:https://glama.ai/mcp/servers

2. Steps to obtain MCP services

The usage methods of these websites are quite similar. Generally speaking:

1. Enter the site and find the service you want on the website (such as fetch webpage scraping, Figma, Gaode Map, etc.).
2. The website page will tell you the sse URL or necessary API Key or other parameters required to use this service.

As shown below: Find the fetch service on the Magic Tower MCP Square. You don't need to provide an additional **API Key**. You just need to obtain the sse URL and then directly configure it in the MCP client for use.

Similarly, for example, enter AIbase (https://mcp.aibase.cn/explore ) and find the Gaode MCP service.

Similarly, there will be a parameter field in the content. What we need is to obtain this field. Note that if the field in this case contains the KEY parameter, it means you need to obtain the key. So, how do you get it?

3. For services that require an API Key, you need to go to the official website of the corresponding service (for example, MCP services such as Figma, Gaode Open Platform, and Exa official website require a Key to be used) to apply for and obtain these API Keys or parameters.

4. Fill the obtained Key into the corresponding input box on the MCP aggregation website.

5. The website will help you generate various import commands or configuration information that can be used in the client according to the information you filled in, usually in Json format.

Therefore, the core process of obtaining MCP services is: Browse MCP aggregation websites → Find services → Obtain API Keys → Generate configurations on the website → Copy the configuration information.

3. Demonstration of obtaining MCP services:

Below, AIbase will specifically show you how to operate by taking the example of obtaining the "Exa Search" MCP service:

1) First, we enter the Magic Tower MCP Square (similarly, you can also choose to visit other MCP aggregation sites to obtain services) and search for the "Exa Search" MCP page.

From the right - hand side, you can see that this service requires obtaining an API key

2) Then, enter the Exa official website: https://exa.ai/ and apply for the free API

3) If you don't have an account, register one. Enter your personal page, click on the API key option, and then click "Create".

4) Edit the name of the KEY and then confirm to create it to obtain the key. Then click the copy button to get the address of the KEY.

5) Enter the obtained keys into the field box on the Magic Tower page and click "Connect".

After connecting, you can get the complete field parameters. Remember to save this obtained field as you'll need it later.

By now, you've completed the operation of obtaining MCP services. The next step is to configure this parameter in the MCP client.

4. How to configure and use MCP services in the client

After getting the MCP services, the next step is to configure them in the MCP client.

With the configuration information of MCP services (usually in Json format), the next step is to import them into the AI client you're using. Currently, there are quite a few clients that support MCP services, including some mainstream chat clients and AI programming IDEs:

Mainstream chat clients:

• Claude
• ChatWise
• Cherry Studio
• DeepChat

AI programming IDEs:

• Trae (a product developed by ByteDance, more user - friendly for Chinese users)
• Cursor
• Windsurf
• Cline - #1 on OpenRouter
• Visual Studio Code - Open Source ("Code - OSS")

Considering that we're targeting ordinary users, AIbase provides detailed configuration methods for several typical clients mentioned above, including Claude, Cherry Studio, ChatWise, Cursor, and Trae. The detailed getting - started tutorials are as follows:

Getting - started tutorial for MCP client: Installation and configuration of Cherry Studio

Getting - started tutorial for MCP client: Installation and configuration of ChatWise

Getting - started tutorial for MCP client: Installation and configuration of Cursor

Getting - started tutorial for MCP client: Installation and configuration of trae

Getting - started tutorial for MCP client: Installation and configuration of Claude

It should be noted that AI IDEs (such as Windsurf and Cursor) may still be a bit complicated for ordinary people. Chatting tools like ChatWise and Cherry Studio may be easier to get started with. Of course, for Chinese users, Trae developed by ByteDance is also a good choice.

Newcomers are advised to choose one of ChatWise, Cherry Studio, and Trae to practice. Once you're familiar with the operation, they're actually quite similar, and you'll gradually get the hang of it.

5. Sharing of practical MCP cases

After covering the theory and teaching the configuration methods, next comes the exciting practical part!

Case 1: Scrape webpage content and rewrite it

1. Basic preparations

Choose an MCP client: chrry studio (for specific configuration operations, you can visit the following getting - started tutorial)

Getting - started tutorial for MCP client: Installation and configuration of Cherry Studio (2)

Obtain an MCP service: fetch (this is a webpage scraping service)

sse configuration url: https://mcp-979bdfce-2a57-47a9.api-inference.modelscope.cn/sse

2. In chrry studio, create an MCP service. According to the configuration page, modify the name, enter the URL, and save it.

3. Use the fetch MCP service to scrape the webpage and rewrite the content according to the instructions

First, enter the dialog box and select the fetch MCP service

Throw in a link with the rewriting instruction:

https://www.chinaz.com/ainews/17600.shtml, summarize the page content in 100 words

The model automatically calls the fetch service and summarizes the webpage content in 100 words.

Case 2: Summarize webpage content and generate audio

The above is the effect of a single MCP running. We can combine multiple MCPs to see what effects can be achieved. For example, combine the MCPs of webpage scraping and audio generation services.

1. Basic preparations

Choose an MCP client: trae (for specific configuration operations, you can visit the following getting - started tutorial)

Getting - started tutorial for MCP client: Installation and configuration of trae (2)

MCP services needed: fetch, minimax

Configuration parameters for fetch:

{
  "mcpServers": {
    "fetch": {
      "url": "<https://mcp-979bdfce-2a57-47a9.api-inference.modelscope.cn/sse>"
    }
  }
}

Configuration parameters for minimax:

{
  "mcpServers": {
    "MiniMax": {
      "command": "uvx",
      "args": [
        "minimax-mcp"
      ],
      "env": {
        "MINIMAX_API_KEY": "Fill in the key applied for from the **minimax official** website,
        "MINIMAX_MCP_BASE_PATH": "The address of the local folder, such as F:/FF",
        "MINIMAX_API_HOST": "<https://api.minimax.chat>",
        "MINIMAX_API_RESOURCE_MODE": "local"
      }
    }
  }
}

Obtain the minimax official API key: Enter the official website > Account Management > Interface Key > Create a new key to obtain it

Link: https://platform.minimaxi.com/user-center/basic-information/interface-key

2. Create an agent

Create an audio assistant agent. Enter the agent prompt in the dialog box and check the MCP services of webpage scraping fetch and audio generation minimax that are needed.

“Summarize the content of the URL into a 100 - word copy and then convert the text into audio”

3. The agent directly calls the MCP services

Next, just select the @audio assistant agent in the trae chat window and throw in a url page. The agent can automatically call the fetch and minimax services. It will first summarize the page content and then generate audio.

For example, when AIbase throws in the link "https://www.chinaz.com/ainews/17477.shtml", the agent automatically summarizes it in 100 words:

“Nano AI releases the MCP Universal Toolbox, pre - installed with over 100 services and 18 API keys, supporting multimodal generation and in - depth search, and simplifying the MCP configuration process. This tool is suitable for fields such as content creation, enterprise automation, and education and is currently in the internal testing phase.”

Finally, a 100 - word copy is automatically generated into audio. The effect is as follows:

t2a_纳米AI发布MCP万_20250424_145044.mp3

Conclusion

The significance of MCP lies in that it transforms AI from the previous role of "just being used by me" into an "active service provider" that can actively call various services and provide customized workflows for you.

Of course, just like any new technology when it first emerges, MCP still has room for improvement in terms of usability. The configuration process may really seem a bit confusing for friends who have never touched the command line. The development of technology is always seeking a balance between "openness and flexibility" and "simplicity and intuitiveness", and this contradiction will gradually be resolved as the ecosystem matures. Maybe in the future, we will really see MCPs that can be directly used in AI assistants just by clicking "Install Tools"!

If you encounter problems during the configuration process, don't give up. You can look back at the tutorial or send the error messages you encounter to AI for help.

Well, this is the end of the beginner's guide to MCP suitable for novice users. I hope it's helpful to you!