Docs

Volcengine Managed Service for Prometheus

Intelligent O&M

ChatOps: Managing VMP by using the MCP service

This topic mainly describes how to deploy the VMP MCP server and use VMP-related features on it.

This topic may contain information about third-party services. The information is for reference only. Volcano Engine does not make a guarantee in any form for the performance and reliability of the third-party tools, and the potential impacts of operations on these tools.

Use scenarios

The VMP MCP server helps operation and maintenance personnel perform metric queries and analysis by using natural language in scenarios such as troubleshooting and data analysis. This improves the efficiency of operation and maintenance, and observability. The following scenarios are supported:

Create a VMP workspace instance in the specified region.

Query the information about all workspaces in the specified region within the current account.

Query the information about all workspaces in the specified region within the current account.

Delete the specified workspace instance in the specified region within the current account.

Delete the specified workspace instance in the specified region within the current account.

Execute a PromQL instant query in the specified VMP workspace.

Execute a PromQL query within a specified time range in the specified VMP workspace.

Query the list of metric names that meet the specified criteria in the specified VMP workspace.

Query all label names of the specified metric in the specified VMP workspace.

Specify time series metadata for the specified label selector.

Background

Introduction to MCP

The Model Context Protocol (MCP) is an open protocol that is used by LLMs to access custom tools and services. An MCP server, such as a VMP MCP server, packages and provides various executable tools. In AI IDEs such as Trae and Cursor, the agent interprets user instructions in natural language, establishes communication between the MCP client and MCP server, and invokes the corresponding tools to complete specific tasks.

Usage after VMP is integrated with MCP

As a dedicated connecting bond between LLMs and VMP, the MCP service enables a natural language-driven intelligent management mode for VMP. The MCP server enables operation and maintenance personnel or business personnel who are not familiar with PromQL to easily perform operations, such as VMP workspace queries and metric retrieval. In addition, it also enables AI to securely and efficiently access and interact with resources across different scenarios. This significantly reduces the difficulty in using VMP.

Prerequisites

A Volcano Engine root user has been created, and the API key of the account has been obtained. For more information, click the link.

tip

When you create an account for the VMP MCP server, we recommend that you conform to the following principles:

Dedicated account: We recommend that you create a dedicated IAM user to perform operations related to the MCP server.

Principle of least privilege: Make sure that your account has permissions on VMP-related APIs, such as CreateCluster and CreateNodePool. Attach the corresponding policies to the account based on the principle of least privilege. For details, see Configuring IAM user permissions.

Procedure

Step 1: Deploy the VMP MCP server

You can deploy the MCP server of the VMP service in your on-premises environment or ModelArk. Select an appropriate method based on your business requirements.

Deploy the server on the on-premises computer

Deploy the server in ModelArk

Deploy the VMP MCP server in your on-premises environment, such as your on-premises computer, by using the AI IDE tool.

Prerequisites

You have deployed the AI IDE platform and related dependencies (Python and uv) in your on-premises environment.

Tool/add-on	Description
AI IDE	In this example, Trae is selected as the AI IDE for subsequent operations. Trae is an AI IDE that supports MCP. You can quickly deploy the MCP server (such as connecting to VMP) by using the built-in MCP marketplace of Trae or through manual configuration. You can go to the TRAE official website to download and install it on your computer.
Python	v3.10 or later. You can visit the official Python website to download and install the tool.
uv	A tool that is used to quickly install and manage Python packages and provides the uvx command-line tool. For more details, see the official guide on uvx. MacOS/Linux curl -LsSf https://astral.sh/uv/install.sh \| sh Windows powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 \| iex"

Procedure

Step 1: Add the VMP MCP server

Add the VMP MCP server from the Volcano Engine LLM ecosystem plaza to your local Trae.

Open the Trae platform installed locally.

Click the settings icon in the upper-right corner of the page.

On the Settings page, click MCP.

Add the VMP MCP server.

Manually add the server

Add the server from the marketplace

Click Add > Add Manually.

Paste the following JSON configurations in the code editor, and replace Your Volcengine access key and Your Volcengine secret key in the code with the actual access key ID and secret access key that correspond to your account.

{
"mcpServers": {
"mcp_server_vmp": {
"command": "uv",
"env": {
"VOLCENGINE_ACCESS_KEY":"Your Volcengine access key",//The access key ID of the Volcano Engine root user.
"VOLCENGINE_SECRET_KEY":"Your Volcengine secret key"//The secret access key of the Volcano Engine root user.
},
"args": [
"--directory",
"/<your local path to mcp-servers>/server/mcp_server_vmp",
"run",
"mcp-server-vmp"
]
}
}
}

{
"mcpServers": {
"mcp_server_vmp": {
"command": "uvx",
"env": {
"VOLCENGINE_ACCESS_KEY":"Your Volcengine access key",//The access key ID of the Volcano Engine root user.
"VOLCENGINE_SECRET_KEY":"Your Volcengine secret key"//The secret access key of the Volcano Engine root user.
},
"args": [
"--from",
"git+https://github.com/volcengine/mcp-server#subdirectory=server/mcp_server_vmp",
"mcp-server-vmp"
]
}
}
}

Environment variables

Environment variable name	Description	Required
VOLCENGINE_ACCESS_KEY	The access key ID of your Volcano Engine root user. To obtain the ID, see Access key usage guide.	Yes
VOLCENGINE_SECRET_KEY	The secret access key of your Volcano Engine root user. To obtain the key, see Access key usage guide.	Yes
VOLCENGINE_REGION	The region of Volcano Engine.	Optional

Click Confirm to save the configurations.

Step 2: Associate the server with an agent

After you add the VMP MCP server, associate it with an agent. In this way, the agent can use the MCP capabilities provided by the VMP service.

Click the settings icon in the upper-right corner of the Trae page.

On the Settings page, click Agents.

Click Create and enter the agent name. In this example, enter VMP service agent.

In the Tool - MCP list, select the configured VMP MCP, and keep other default settings.

Click Confirm to complete the configuration.

Step 2: Use the server

Use the server on the on-premises computer

Use the server in ModelArk

After you create the MCP server, click Use now. In the dialog box with the agent, enter instructions in natural language as needed to interact with the agent. This topic uses the instruction Query the information about all workspaces in the specified region within the current account as an example.

Enter Query the information about all workspaces in the specified region within the current account in the dialog box with the agent.

Press Enter to view the answer returned by the agent. If the agent indicates that more information is required, enter the related information to view the feedback result.

tip

After the agent receives the instruction, it analyzes and determines which specific tool provided by the VMP MCP server needs to be used.

The agent automatically generates API request parameters and calls the tool provided by the VMP MCP server by using MCP.

After the MCP server executes the query, it returns the result to the agent.

The agent organizes the result and presents it to you in an easy-to-understand format, such as charts and text.

Last updated: 2026.10.10 11:04:51