MCP Connection Guide

This document provides a getting-started guide for connecting to and using MCP (Model Control Plane) on the SERVICEME platform. Whether you are a business user or a technical user, this guide will help you quickly get started with MCP.

What is MCP
MCP Connection Types
Method 1: Use an MCP Published by Your Organization
Method 2: Create and Connect an External MCP Service
Method 3: Publish an Agent as MCP

What is MCP

MCP (Model Control Plane) is the central module in AI Studio used to connect AI models with external systems. It enables Agents to:

🔌 Connect to external services: call third-party APIs and enterprise system interfaces
🛠️ Extend capabilities: add computing, search, data analysis, and other capabilities
🤝 Cross-system collaboration: enable data interaction across multiple systems
🔒 Security governance: centrally manage permissions and access control

Simply put, MCP is the key component that turns AI from "able to talk" into "able to act".

Typical MCP Application Scenarios

Computing services: provide Agents with mathematical capabilities (addition, subtraction, multiplication, division, scientific computing, etc.)
Search services: connect to web search engines, academic paper retrieval, and enterprise knowledge base search
Data analysis: generate charts, execute data queries, and produce analytical reports
Tool integration: integrate map services, translation services, file processing tools, etc.
Business systems: connect to internal enterprise systems such as CRM, ERP, and databases

MCP Connection Types

Depending on the source of the MCP and how it is configured, the following connection types are available:

Connection Type	Applicable Scenario	Difficulty	Configuration Requirement	Recommended Users
Organization MCP	Use services already configured by the team	⭐ Easy	No configuration required, just select directly	All users
External Service MCP	Connect to third-party APIs or self-built systems	⭐⭐⭐ Medium	JSON configuration required	Technical users
Agent to MCP	Publish an existing Agent as MCP	⭐⭐ Easy	Publishing permission required	Authorized users

Selection recommendations:

If this is your first time using MCP, it is recommended to start with Organization MCP
If you need to connect to a specific external service, choose External Service MCP
If you want to share a mature Agent with your team, use Agent to MCP

Method 1: Use an MCP Published by Your Organization

This is the simplest method and is suitable for all users to get started quickly.

Applicable Scenarios

Your team has already configured commonly used MCP services
You only need to use ready-made capabilities without configuring them yourself
Quickly add specific capabilities to an Agent

Steps

1. Go to the Agent configuration page

In AI Studio, open the Agent to which you want to add MCP capabilities
Click the "Configure" or "Edit" button in the upper-right corner
Find the "MCP Services" section on the configuration page

2. Select an MCP service

Click the "Add MCP" or "Select MCP" button
Browse the list of MCPs published within the organization
You can filter and search in the following ways:
- By category: computing services, search services, data analysis, tool integration, etc.
- By name: search for a specific MCP name
- By function: view the specific tools provided by the MCP

Search tips:

Use the search box to quickly locate a specific MCP
Check the MCP description to understand its functions and use cases
Review the list of tools provided by the MCP to confirm whether it meets your needs

3. Enable MCP tools

After selecting the required MCP, the system will display all tools provided by that MCP (Tools)
Check the specific tools you want to enable
- For example, in the calculator MCP, you can choose:
  - add (addition)
  - subtract (subtraction)
  - multiply (multiplication)
  - divide (division)
Click "Confirm" to add

Notes:

Enable only the tools you need to avoid adding too many unnecessary capabilities to the Agent
Some MCPs may require administrator authorization before they can be used
Review the parameter requirements of the tools to understand how to use them correctly

4. Save and publish

After completing the MCP configuration, click the "Save" button in the upper-right corner
Click "Publish" to make the changes take effect
The Agent can now use the capabilities of the selected MCP

Usage Examples

After configuration is complete, when chatting with the Agent, the Agent will automatically call the corresponding MCP tools:

Example 1: Using the calculator MCP

User: Help me calculate 1234 + 5678
Agent: [Calls the add tool of the calculator MCP] 
       1234 + 5678 = 6912

Example 2: Using the search MCP

User: Search for the latest artificial intelligence news
Agent: [Calls the search tool of the search MCP]
       Based on the search results, here is the latest artificial intelligence news...

Example 3: Using the weather MCP

User: What's the weather like in Beijing today?
Agent: [Calls the getCurrentWeather tool of the weather MCP]
       Beijing is cloudy today, with a temperature of 15-25°C and good air quality.

The system will display an MCP invocation indicator, showing that the Agent used external tools to complete the task.

Frequently Asked Questions

Q: How do I know which MCPs are available in the organization?
A: You can browse all MCPs published by the organization in the MCP selection interface. It is recommended to ask your team administrator or check internal documentation.

Q: Why can't I see some MCPs?
A: It may be due to permission restrictions. Please contact the administrator to assign the appropriate access permissions.

Q: Can multiple MCPs be used at the same time?
A: Yes. An Agent can use multiple MCPs simultaneously to achieve more complex functional combinations.

Q: How do I remove an MCP that has already been added?
A: Go to the Agent configuration page, find the corresponding MCP in the MCP Services section, and click "Remove" or uncheck it.

Method 2: Create and Connect an External MCP Service

This is suitable for users with a technical background and allows connection to custom external services or third-party APIs.

Applicable Scenarios

You need to connect to internally built systems within the organization
You want to use a specific third-party API service
Existing MCPs cannot meet special requirements

Prerequisites

Before you begin, please make sure you have:

✅ Basic technical knowledge

Understand basic JSON format and syntax
Be familiar with basic command-line concepts (such as commands like node and python)

✅ Service information

Have the API documentation for the service you want to connect to
Know the service access address and port
Understand the service invocation method and parameter requirements

✅ Authentication credentials

Have access permission to the service
Possess an API Key, Token, or other authentication information
Confirm the authentication method (such as OAuth, API Key, etc.)

Steps

1. Create a new MCP

In the left navigation bar of AI Studio, click "MCP"
Click the "Create" button in the upper-right corner
Enter the MCP creation interface

2. Fill in the basic information

MCP Name （Required）

Set a clear and recognizable name for the MCP
Recommended format: [service-type]-[function-description]
Examples:
- weather-api - weather query service
- crm-system - CRM system integration
- translation-service - translation service
- database-query - database query tool
Limit: within 50 characters

MCP Avatar （Optional）

Select one from the default avatars provided by the system
Choosing an icon related to the service type helps with identification
Custom upload is not supported at this time

MCP Description （Required）

Describe in detail the functions, purpose, and application scenarios of the MCP
Explain what service it connects to and what capabilities it provides
Help other users understand when to use this MCP

Example:

Connects to a weather query API and provides real-time weather queries,
weather forecasts, air quality queries, and other functions. Supports major cities worldwide.
Suitable for scenarios such as customer service and travel planning that require weather information.

Limit: within 200 characters

MCP Category （Required）

Select the most appropriate category for easier management and search later
Available categories:
- Computing services: mathematical computation, data processing, etc.
- Search services: information retrieval, web search, etc.
- Data analysis: report generation, data visualization, etc.
- Tool integration: maps, translation, file processing, etc.
- Other: services that do not belong to the above categories

3. Configure the MCP service

This is the most critical step and requires filling in the service configuration in JSON format.

Basic configuration structure:

{
  "mcpServers": {
    "服务名称": {
      "command": "启动命令",
      "args": ["参数1", "参数2"],
      "env": {
        "API_KEY": "你的API密钥",
        "BASE_URL": "服务基础URL"
      }
    }
  }
}

Configuration field descriptions:

Field	Required	Description	Example
`mcpServers`	✅	Root object of the MCP service	-
`服务名称`	✅	Define a unique identifier for the service	"calculator", "weather"
`command`	✅	Command to start the MCP service	"node", "python", "npx"
`args`	✅	Array of parameters passed to the command	`["script.js", "--port", "3000"]`
`env`	⚪	Environment variable configuration	`{"API_KEY": "xxx"}`

Simple example - calculator service:

{
  "mcpServers": {
    "calculator": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-calculator"
      ]
    }
  }
}

The meaning of this configuration:

Create an MCP service named "calculator"
Run it using the npx command
The parameter -y means automatically confirm installation
The parameter @modelcontextprotocol/server-calculator is the npm package to run

4. Save the configuration

Carefully check whether the JSON format is correct
- Make sure all commas, quotation marks, and brackets match
- Confirm there is no extra or missing punctuation
Click the "Save" button
The system will validate the JSON format and prompt you to make corrections if there are errors

Common formatting errors:

❌ Missing comma: "key1": "value1" "key2": "value2"
✅ Correct: "key1": "value1", "key2": "value2"
❌ Extra comma: ["item1", "item2",]
✅ Correct: ["item1", "item2"]

5. Test the MCP

After saving, it is recommended to test immediately whether the MCP is working properly:

Click the "Test" button
Check whether the "MCP Tools" list is displayed on the right side
Confirm that all expected tools have been loaded successfully

If the test fails, please check:

Whether the JSON configuration is correct
Whether the command and parameters are valid
Whether the network connection is working properly
Whether the API key is correct (if required)

Configuration Examples

To help you better understand the configuration, here are several commonly used examples:

Example 1: MCP using an npm package

{
  "mcpServers": {
    "calculator": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-calculator"]
    }
  }
}

Example 2: MCP with environment variables

{
  "mcpServers": {
    "weather": {
      "command": "node",
      "args": ["path/to/weather-service.js"],
      "env": {
        "API_KEY": "your_api_key_here",
        "BASE_URL": "https://api.weatherapi.com/v1"
      }
    }
  }
}

Example 3: MCP for a Python service

{
  "mcpServers": {
    "data-analysis": {
      "command": "python",
      "args": ["analysis_server.py"],
      "env": {
        "DATA_SOURCE": "database",
        "LOG_LEVEL": "info"
      }
    }
  }
}

Method 3: Publish an Agent as MCP

This is suitable for users who already have a mature Agent. You can package its capabilities as an MCP so that other members within the organization can call it directly without repeated configuration.

Applicable Scenarios

You have already created a fully functional Agent and want to share its capabilities with the team
You need to standardize specific agent capabilities (such as automated reporting, knowledge retrieval, process approval, etc.)
You want other Agents to be able to directly call the current Agent's functions

Prerequisites

Before you begin, please make sure:

✅ A personal Agent has been created and fully configured
✅ The Agent's functions have been thoroughly tested and run stably
✅ If you need to publish it as an organization-level MCP, you must have permission to make the Agent public as an organizational agent

Steps

1. Go to the Agent configuration page and click Publish as MCP

Open the Agent you want to publish in AI Asset
Click the MCP icon button (plug-shaped icon) in the upper-right corner
Select "Publish as MCP" from the dropdown menu

2. Confirm publishing

The system will pop up a confirmation dialog to confirm the publishing information
After confirmation, the Agent will automatically generate a corresponding MCP and it will appear on the AI Asset → My → MCP page
At this point, the MCP status is Available, but it is for personal use only

3. Make it public as an organization MCP (optional)

If you want other members in the organization to also use this MCP, you need to make it public:

Go to the AI Asset → My → MCP page
Find the MCP you just published and click to enter the details page
Click the "Public" button in the upper-right corner
After confirmation, the MCP will appear in the organization's MCP list for all members to use

Usage Instructions

After successful publishing, other users can find and use this MCP under the Org tab by following the steps in Method 1.

Notes:

After an Agent is published as an MCP, it will not affect the normal use of the original Agent
If the configuration of the original Agent changes (such as updates to the Prompt, tools, etc.), it is recommended to republish it to ensure the MCP stays synchronized with the Agent
After making it public as an organization MCP, please ensure the Agent is sufficiently stable to avoid affecting other members' usage

Frequently Asked Questions

Q: After publishing as MCP, can the original Agent still be used normally?
A: Yes. Publishing as MCP is an independent operation and will not affect any functions of the original Agent.

Q: How do I cancel public access and change the MCP back to personal-only visibility?
A: Go to the MCP details page, click the settings next to the "Public" button, and change the visibility scope back to private.

Q: After the Agent is updated, will the published MCP be synchronized automatically?
A: It will not be synchronized automatically. You need to perform the "Publish as MCP" operation again to update the MCP content.

Table of Contents​

What is MCP​

Typical MCP Application Scenarios​

MCP Connection Types​

Method 1: Use an MCP Published by Your Organization​

Applicable Scenarios​

Steps​

1. Go to the Agent configuration page​

2. Select an MCP service​

3. Enable MCP tools​

4. Save and publish​

Usage Examples​

Frequently Asked Questions​

Method 2: Create and Connect an External MCP Service​

Applicable Scenarios​

Prerequisites​

Steps​

1. Create a new MCP​

2. Fill in the basic information​

3. Configure the MCP service​

4. Save the configuration​

5. Test the MCP​

Configuration Examples​

Method 3: Publish an Agent as MCP​

Applicable Scenarios​

Prerequisites​

Steps​

1. Go to the Agent configuration page and click Publish as MCP​

2. Confirm publishing​

3. Make it public as an organization MCP (optional)​

Usage Instructions​

Frequently Asked Questions​

Table of Contents

What is MCP

Typical MCP Application Scenarios

MCP Connection Types

Method 1: Use an MCP Published by Your Organization

Applicable Scenarios

Steps

1. Go to the Agent configuration page

2. Select an MCP service

3. Enable MCP tools

4. Save and publish

Usage Examples

Frequently Asked Questions

Method 2: Create and Connect an External MCP Service

Applicable Scenarios

Prerequisites

Steps

1. Create a new MCP

2. Fill in the basic information

3. Configure the MCP service

4. Save the configuration

5. Test the MCP

Configuration Examples

Method 3: Publish an Agent as MCP

Applicable Scenarios

Prerequisites

Steps

1. Go to the Agent configuration page and click Publish as MCP

2. Confirm publishing

3. Make it public as an organization MCP (optional)

Usage Instructions

Frequently Asked Questions