MCP Service Management

Download

Focus Mode

Font Size

Last updated: 2026-09-22 18:47:45

AI-Translated

Scenarios
With the widespread adoption of large model applications, an increasing number of business systems aim to expose their capabilities to AI Agents via the Model Context Protocol (MCP). The AI Gateway serves as a unified MCP access layer. It shields you from underlying differences and provides a unified access point, protocol conversion, Tool management, and multi-backend aggregation capabilities. This enables you to securely and conveniently expose your backend services externally in a standard MCP format.
The AI Gateway supports the following two MCP service types to accommodate different access scenarios:
Standard MCP Service: This service is suitable for scenarios where the backend has already implemented the MCP protocol based on the official MCP SDK, or has registered with the TSF Nacos MCP Registry. The gateway acts as a pass-through proxy, forwarding MCP requests from clients to the backend unchanged. Tool metadata is automatically synchronized by the backend and appears as read-only in the console.
HTTP-to-MCP Service: This service is suitable for scenarios requiring the rapid upgrade of existing REST/HTTP APIs to MCP capabilities. The gateway handles the conversion between HTTP and MCP protocols. Tools are defined by you as needed, supporting manual creation or batch import based on OpenAPI specification files. This service type also includes the Virtual MCP Server mode, which can aggregate interfaces from multiple different systems and expose them as a single MCP Server externally.
This document describes how to configure service sources and create backend services in the AI Gateway, as well as how to add, view, and edit MCP services.
Operation Steps
Configuring a Service Source
To complete the service source configuration, see the Service Management - Service Source section.
Configuring Backend Services
To create an HTTP-to-MCP Service (Domain Name/IP Mode) or a Standard MCP Service (Domain Name/IP Mode), you must first create a service under the Service Management > Service tab to serve as the instance associated with the MCP service. See the Service Management - Service section to complete the service configuration.
Creating an MCP Service
1. Log in to the Microservices Platform console. In the left sidebar, click Cloud Native Intelligent Gateway > Instance List.
2. On the instance list page, click the ID of the gateway instance you want to configure to go to its basic information page.
3. In the left sidebar, click MCP Management. Above the MCP service list, click New.
4. In the "New MCP Service" window, complete the configuration for the first step, "Basic Information".
Parameter
Required
Description
MCP service name
Yes
Enter the unique identifier for the MCP service. This name will be used as part of the access path (/mcpservers/<MCP service name>/mcp). It can contain up to 60 characters, supporting uppercase and lowercase English letters, digits, and separators ("-", "_"). It cannot start with a digit or separator, and cannot end with a separator. It must be unique within the same gateway instance and cannot be modified after creation.
Display name
Yes
Enter the service display name, which is used for display in the console list and details page. It can contain up to 60 characters, supporting uppercase and lowercase letters in both Chinese and English, digits, and separators ("-", "_"). It cannot start with a digit or separator, and cannot end with a separator.
Service type
Yes
Select the MCP service type:
Standard MCP Service: The MCP protocol is implemented by the backend, and the gateway acts as a pass-through proxy. Tools are automatically synchronized and are read-only.
HTTP-to-MCP Service: For existing HTTP services, the gateway handles protocol conversion and includes the Virtual MCP Server mode. Tools are configured by users.
Request protocol
Yes
The MCP transport protocol used when a client calls the gateway. The current version is fixed to Streamable HTTP.
Description
No
The description of this service, which facilitates subsequent management. It can contain up to 200 characters.
Host passthrough
No
When the feature is enabled, the gateway retains the Host header from the original client request and forwards it to the backend MCP Server. When the feature is disabled, the gateway rewrites it to the backend address.
5. After completing the basic information, click Next to go to the "Backend Service Configuration" step. The system will display different backend configuration items based on the Service Type selected in "Basic Information". There are four scenarios in total.
Scenario 1: Standard MCP Service + MCP Registry
This applies when the backend MCP Server is registered with TSF Nacos, and the gateway automatically synchronizes service addresses and Tool metadata from the registry. In the Backend Type section, select MCP Registry and complete the configuration by referring to the table below.
Parameter
Required
Description
Service Protocol
Yes
Select the protocol for the backend MCP Server. The default is HTTP, and HTTPS is optional.
Service Source
Yes
Select the service source of the MCP registry type from the drop-down list. If not created, click New Service Source to quickly jump to the creation process.
Namespace
Yes
Enter the Nacos namespace. Up to 128 characters. Supports only uppercase and lowercase letters in English, digits, _, and -.
MCP service
Yes
Enter the service identifier for the target MCP Server. Up to 128 characters. Supports only uppercase and lowercase letters in English, digits, _, and -.
MCP Endpoint
Yes
Enter the Endpoint path for the MCP protocol. Up to 64 characters. Starts with /. Supports only uppercase and lowercase letters in English, digits, /, _, and -.
Service Group
Yes
Enter the Nacos service group. Up to 128 characters. Supports only uppercase and lowercase letters in English, digits, ., and -.
Service Name
Yes
Enter the Nacos service name. Up to 128 characters. Supports only uppercase and lowercase letters in English, digits, ., and -.
Timeout Time
Yes
Timeout for gateway requests to the backend. Default value: 3000 ms. Value range: 0 to 60000.
Number of retries
Yes
Number of retries after request failures. Default value: 3. Value range: 0 to 10.
Health Check
No
Whether to enable MCP Server health check. If health check is required, turn on the button and configure health check parameters:
Check Type: HTTP/HTTPS
Check Interval: Greater than 5 seconds (required)
Timeout: Greater than 1 second (required)
Failure Threshold: Greater than 1 time (required)
Recovery Threshold: Greater than 1 time (required)
Probe Path: Configure it yourself. The default path is /health (required).
Probe Method: GET/POST
Scenario 2: Standard MCP Service + Domain Name/IP address
This applies to scenarios where the backend MCP Server does not use a registry and requires manual specification of the service address. In the Backend Type section, select Domain Name/IP.
Parameter
Required
Description
Service Protocol
Yes
Select HTTP or HTTPS.
Service Address
Yes
Enter the IP address or domain name of the backend MCP Server, for example, 10.0.0.1 or mcp-server.example.com.
Service port
Yes
Enter the port of the backend service. Any port from 1 to 65535 is supported.
MCP Endpoint
Yes
Enter the Endpoint path for the MCP protocol. Up to 64 characters. Starts with /. Supports only uppercase and lowercase letters in English, digits, /, _, and -.
Timeout Time
Yes
Default value: 3000 ms. Value range: 0 to 60000.
Number of retries
Yes
Default value: 3. Value range: 0 to 10.
Health Check
No
Whether to enable MCP Server health check. If health check is required, turn on the button and configure health check parameters.
Check Type: HTTP/HTTPS
Check Interval: Greater than 5 seconds (required)
Timeout: Greater than 1 second (required)
Failure Threshold: Greater than 1 time (required)
Recovery Threshold: Greater than 1 time (required)
Probe Path: Configure it yourself. The default path is /health (required).
Probe Method: GET/POST
Scenario 3: Standard MCP Service + TKE
Automatically discover Services from TKE clusters, with no awareness of Pod drift.
Parameter
Required
Description
Service Protocol
Yes
Select HTTP or HTTPS.
Service Source
Yes
Select configured TKE source.
Service port
Yes
Enter the namespace of the target TKE service.
K8s Service
Yes
Target K8s service. In multi-machine multi-GPU scenarios, select the master node service.
Service port
Yes
Port of the backend service. Any port from 1 to 65535 is supported.
MCP Endpoint
Yes
Enter the Endpoint path for the MCP protocol. Up to 64 characters. Starts with /. Supports only uppercase and lowercase letters in English, digits, /, _, and -.
Timeout Time
Yes
Default value: 3000 ms. Value range: 0 to 60000.
Number of retries
Yes
Default value: 3. Value range: 0 to 10.
Health Check
No
Whether to enable MCP Server health check. If health check is required, turn on the button and configure health check parameters.
Check Type: HTTP/HTTPS
Check Interval: Greater than 5 seconds (required)
Timeout: Greater than 1 second (required)
Failure Threshold: Greater than 1 time (required)
Recovery Threshold: Greater than 1 time (required)
Probe Path: Configure it yourself. The default path is /health (required).
Probe Method: GET/POST
Scenario 4: HTTP to MCP Service + Domain Name/IP address
This method is suitable for connecting to a single backend HTTP service, where all Tools share the same backend service instance. In the Backend Type section, select Domain Name/IP and complete the configuration by referring to the table below.
Parameter
Required
Description
Backend service
Yes
Select a backend service that has been created under Service Management > Services from the drop-down list. If not created, click New Service to quickly jump to the creation process, or click Refresh to reload the drop-down data.
Scenario 5: HTTP to MCP Service + Virtual MCP Server
This method is suitable for scenarios where multiple different HTTP APIs are aggregated and exposed as a single MCP Server. The gateway itself acts as the MCP Server, and each Tool is configured with a complete backend URL independently.
Scenario 6: HTTP to MCP Service + Standard Registry
This method is suitable for quickly wrapping registry services into MCP services.
Parameter
Required
Description
Service Protocol
Yes
Select the protocol for the backend MCP Server. The default is HTTP, and HTTPS is optional.
Registry Type
Yes
TSF Nacos,TSF PolarisMesh
Service Source
Yes
Select the service source of the common registry type from the drop-down list. If not created, click New Service Source to quickly jump to the creation process.
Namespace
Yes
Enter the Nacos/PolarisMesh namespace. Up to 128 characters. Supports only uppercase and lowercase letters in English, digits, _, and -.
Service
Yes
Enter the service identifier for the target MCP Server. Up to 128 characters. Supports only uppercase and lowercase letters in English, digits, _, and -.
Service Group
Yes
Enter the Nacos service group. Up to 128 characters. Supports only uppercase and lowercase letters in English, digits, ., and -.
Timeout Time
Yes
Timeout for gateway requests to the backend. Default value: 3000 ms. Value range: 0 to 60000.
Number of retries
Yes
Number of retries after request failures. Default value: 3. Value range: 0 to 10.
Health Check
No
Whether to enable MCP Server health check. If health check is required, turn on the button and configure health check parameters:
Check Type: HTTP/HTTPS
Check Interval: Greater than 5 seconds (required)
Timeout: Greater than 1 second (required)
Failure Threshold: Greater than 1 time (required)
Recovery Threshold: Greater than 1 time (required)
Probe Path: Configure it yourself. The default path is /health (required).
Probe Method: GET/POST
Scenario 7: HTTP to MCP Service + Container Service
Applicable to automatic Service discovery from TKE clusters, with no awareness of Pod drift. The configuration method is the same as Scenario 3.
6. After completing the configuration, click OK to create the MCP service. Once the creation is successful, the newly added MCP service will appear in the service list. Click its ID to view detailed service information.
For Standard MCP Services, Tools are automatically synchronized by the backend MCP Server and can be viewed on the Tools Management tab of the service details page.
For HTTP-to-MCP Services, go to the Tools Management tab on the service details page to manually create a Tool or import them in batches via OpenAPI. For details, see MCP Tools Management.
Viewing MCP Service Details
In the MCP service list, click the ID/Name of the target service to go to the service details page. The details page contains two tabs: Basic Information and Tools Management.
The Basic Information tab displays the following three sections:
Basic Information Section: This section displays metadata such as the MCP service ID, MCP service name, service display name, service type, request protocol, creation time, modification time, and description. Click Edit in the upper-right corner to modify variable fields like the service display name and description.
Invocation Method Section: This section displays the mcpServers JSON configuration template for AI Agents. It supports one-click copying and pasting into the configuration files of MCP clients such as Cursor, CodeBuddy, and Claude Desktop. The private network invocation address format is as follows:
{
  "mcpServers": {
    "<MCP Service Name>": {
      "url": "http://<Gateway IP>/mcpservers/<MCP Service Name>/mcp"
    }
  },
  "transportType": "streamable-http"
}
The gateway access path is fixed as /mcpservers/<MCP Service Name>/mcp and is automatically generated by the gateway based on the instance IP address and service name. To access via the public network, enable the public network CLB in the Basic Information > Network Configuration of the AI Gateway instance. After it is enabled, the public network access address will be displayed in the Invocation Method section.
Backend Service Configuration Section: This section displays the backend type and its corresponding fields. In Virtual MCP Server mode, the service protocol is displayed as -, indicating that there is no unified service-level backend protocol. Each Tool is specified independently during its creation.
The Tools Management tab is used to manage the Tools under this service. For details, see MCP Tools Management.
Editing an MCP Service
In the MCP service list, locate the target service and click Edit under its Actions column to go to the Edit MCP Service window.
In the first step, "Basic Information", fields such as MCP Service ID, MCP Service Name, Service Type, and Request Protocol are read-only. Only the Service Display Name and Description are editable. After clicking Next, you can modify the fields in the backend service configuration, such as the service address, service port, timeout period, retry quantity, and backend service. After completing the modifications, click OK to save and apply the changes.
Must-Knows
Standard MCP Service (MCP Registry Mode) relies on the version features of TSF Nacos 3.x. Please confirm the version of the selected Nacos instance.
For an HTTP-to-MCP Service (Domain Name/IP Mode), you must first create a service under the Service Management > Service tab, and then reference it when creating a new MCP service. When creating an MCP service in Virtual MCP Server Mode, you do not need to configure a backend address.

Help and Support

Was this page helpful?

You can also Contact sales or Submit a Ticket for help.

Help us improve! Rate your documentation experience in 5 mins.

Feedback

Parameter	Required	Description
MCP service name	Yes	Enter the unique identifier for the MCP service. This name will be used as part of the access path (`/mcpservers/<MCP service name>/mcp`). It can contain up to 60 characters, supporting uppercase and lowercase English letters, digits, and separators ("-", "_"). It cannot start with a digit or separator, and cannot end with a separator. It must be unique within the same gateway instance and cannot be modified after creation.
Display name	Yes	Enter the service display name, which is used for display in the console list and details page. It can contain up to 60 characters, supporting uppercase and lowercase letters in both Chinese and English, digits, and separators ("-", "_"). It cannot start with a digit or separator, and cannot end with a separator.
Service type	Yes	Select the MCP service type: Standard MCP Service: The MCP protocol is implemented by the backend, and the gateway acts as a pass-through proxy. Tools are automatically synchronized and are read-only. HTTP-to-MCP Service: For existing HTTP services, the gateway handles protocol conversion and includes the Virtual MCP Server mode. Tools are configured by users.
Request protocol	Yes	The MCP transport protocol used when a client calls the gateway. The current version is fixed to Streamable HTTP.
Description	No	The description of this service, which facilitates subsequent management. It can contain up to 200 characters.
Host passthrough	No	When the feature is enabled, the gateway retains the Host header from the original client request and forwards it to the backend MCP Server. When the feature is disabled, the gateway rewrites it to the backend address.

tencent cloud

Cloud Native Intelligent Gateway

MCP Service Management

Scenarios

Operation Steps

Configuring a Service Source

Configuring Backend Services

Creating an MCP Service

Viewing MCP Service Details

Editing an MCP Service

Must-Knows

Help and Support