Skip to main content

MCP User Guide

Version Information: This document provides a comprehensive introduction to the Model Context Protocol (MCP) and practical usage methods, helping developers quickly get started and integrate it into various AI applications.

What is MCP?​

MCP (Model Context Protocol) is an open-source protocol designed to provide dynamic and rich context information to large language models (LLMs) in a standardized manner. Through MCP, models can flexibly access external data sources and services, significantly enhancing their understanding and reasoning capabilities.

Analogous Understanding​

Think of MCP as the "USB drive" of the AI world:
Just as a USB drive supports flexible reading of various types of data files, the MCP Server can mount various context plugins, which language models can call as needed, enabling "plug-and-play" functionality.

Comparison with Function Tools​

FeatureFunction ToolMCP
Target AccessSpecific function or APIAny structured context resource
Model InteractionFunction CallContext Injection
Modularity & ExtensibilityLowHigh, supporting plugin-based and multi-source access
Applicable ScenariosClearly defined tasksIntelligent applications requiring complex contextual awareness

Use Cases​

MCP can be widely applied to intelligent agent systems that need dynamic context awareness. Below are some typical use cases:

1. Intelligent Travel Planning Assistant​

  • Traditional Approach: Requires separate integrations for calendar, email, airline booking APIs, each with its own authentication, data extraction, and error handling logic.
  • With MCP: The AI assistant accesses calendar, email, and booking services through the MCP Server with a single integration, simplifying the process.

2. Advanced Smart IDE​

  • Traditional Approach: Requires manual integration of multiple components such as file system, Git, package manager, and code documentation.
  • With MCP: The IDE accesses all components through a unified protocol, enhancing features like code generation and auto-completion.

3. Data Analysis & Insights Platform​

  • Traditional Approach: Requires maintaining connections for databases, data cleaning workflows, and visualization tools.
  • With MCP: The model automatically accesses multiple data sources and tools, enabling unified data interaction and context processing.

Quick Start: Creating MCP Server​

Step 1: Provide Basic Information​

When creating an MCP Server, provide a unique name and a brief description for identification and organization.

Step 2: Choose Configuration Method​

You can choose one of the following methods to configure the MCP Server:

Add the MCP Server using a JSON structure. You can directly obtain configuration examples from the following MCP navigation platforms:

Sample Configuration:

"mcpServers": {
  "fetch": {
    "command": "uvx",
    "args": ["mcp-server-fetch"]
  }
}

2.2 Manual Configuration Mode​

Manual configuration supports two modes:

2.2.1 Local Execution (STDIO Mode)​

This mode is suitable for running the MCP Server on a local machine. You need to specify the execution command and optional environment variables.

  • Execution Command Example:
npx -y tavily-mcp@0.1.4
  • Environment Variables (Optional):
TAVILY_API_KEY=xxxxxxx
2.2.2 Remote Service (SSE Mode)​

This mode is suitable for remote service calls to interact with the MCP Server. You need to provide the service URL and optional request headers.

  • Sample URL:
https://mcp.sentry.dev/sse
  • Headers Example:
Authorization: Bearer sk-xxxxxxx

Step 3: Save and Validate Configuration​

After clicking the "Confirm" button, the system will automatically validate the command or URL for legality. After validation, the MCP Server will remain in the off state by default, and you can manually start it as needed.

Configuration Example

Built-in Service Plugins​

The MCP system comes with multiple high-value plugins, ready to use:

Plugin NameDescription
Playwright AutomationA Model Context Protocol (MCP) server that provides browser automation capabilities using Playwright. This server enables LLMs to interact with web pages through structured accessibility snapshots, bypassing the need for screenshots or visually-tuned models.
fetchFetch is a Model Context Protocol (MCP) server designed for web content fetching and conversion, allowing Large Language Models (LLMs) to retrieve and process content from web pages by converting HTML into markdown for easier consumption.
Pages DeployAbility to quickly deploy HTML content to EdgeOne Pages and generate publicly accessible links. This allows you to instantly preview and share AI-generated web content.
TavilyThe AI assistant can seamlessly integrate with Tavily’s advanced search and data extraction capabilities. This integration allows the AI model to access network information in real-time and is equipped with complex filtering options and specific domain search functions.
Desktop CommanderUse AI to search, update, manage local files and run terminal commands.

Check MCP Server Availability​

After adding an MCP Server, you can trigger the check process by clicking the "Check" button in the top-right corner:

  1. Pull dependencies or remotely verify the configuration's validity;
  2. Enable the service after successful validation;
  3. Display all the tools attached to the MCP Server with detailed information.

⚠ The check process for STDIO mode may take longer, depending on the size of dependencies.

MCP Server Check Interface

Use MCP Server in Bot​

On the Bot page, you can add an MCP Server by following these steps:

  1. Click "Add Tool";
  2. Select the desired MCP Server (multi-selection supported);
  3. Confirm and save the configuration.

Binding Example
Usage Example

Network Optimization Recommendations (for China)​

To improve the experience when using MCP in China, we recommend setting the following environment variables:

npx Mirror Configuration:​

NPM_CONFIG_REGISTRY=[{ name: '淘宝 NPM Mirror', url: 'https://registry.npmmirror.com' }]

uv / uvx Mirror Configuration (for PyPI):​

PIP_INDEX_URL=[{ name: '清华大学', url: 'https://pypi.tuna.tsinghua.edu.cn/simple' },{ name: '阿里云', url: 'http://mirrors.aliyun.com/pypi/simple/' },{ name: '中国科学技术大学', url: 'https://mirrors.ustc.edu.cn/pypi/simple/' },{ name: '华为云', url: 'https://repo.huaweicloud.com/repository/pypi/simple/' },{ name: '腾讯云', url: 'https://mirrors.cloud.tencent.com/pypi/simple/' }]

mcp_image_website_config

Summary​

MCP is not just a protocol, but a context connection standard for future AI applications. With its unified access method and modular plugin system, developers can quickly build intelligent applications with context-awareness, enhancing model interaction efficiency and overall system intelligence.