MCP Server auto-registers and Discovery Manual

By developing an MCP Server using the Spring AI Alibaba framework or the Nacos MCP Wrapper Python, the server can be dynamically registered into Nacos after startup and supports the following capabilities:

Dynamic MCP Server Management: Add, delete, update, and query service information via the MCP service list
Dynamic Description Updates: Tool descriptions and parameter definitions support runtime hot updates without requiring a service restart
Dynamic On/Off for MCP Server Tools: Enable or disable tools at runtime without restarting the service
Full-Stack Integration: Service registration information is automatically synchronized to the Nacos configuration center and service discovery module, adapting to AI Agent invocation requirements

Services registered in the Nacos MCP Registry can be discovered and invoked through the Spring AI Alibaba framework or the Nacos MCP Router, or integrated with the Higress gateway to achieve full-chain integration.

Auto-register

Automatic Registration of MCP Server

Developing an MCP Server Using Spring AI Alibaba Nacos MCP Framework

1. Dependency Configuration

<dependency>
    <groupId>com.alibaba.cloud.ai</groupId>
    <artifactId>spring-ai-alibaba-starter-mcp-registry</artifactId>
    <version>{version >= 1.0.0.3}</version>
</dependency>

<!-- MCP Server (WebMVC) -->
<dependency>
    <groupId>org.springframework.ai</groupId>
    <artifactId>spring-ai-starter-mcp-server-webmvc</artifactId>
    <version>{version >= 1.0.0}</version>
</dependency>

2. Service Definition (Example)

@Service
public class WeatherService {

  @Tool(description = "Get weather information by city name")
  public String getWeather(@ToolParam(description = "City name") String cityName) {
    return "Sunny in " + cityName;
  }
}

3. Auto-registration Configuration

spring:
  application:
    name: mcp-nacos-registry-example
  ai:
    mcp:
      server:
        name: webmvc-mcp-server   # MCP service name
        version: 1.0.0            # Service version
        type: SYNC                # Invocation type: SYNC (synchronous) or ASYNC (asynchronous)
        instructions: "This mcp server provides time information tools and resources"
    alibaba:
      mcp:
        nacos:
          server-addr:          # Replace with your Nacos address
          namespace: public    # Nacos namespace ID (default is 'public')
          username:           # Open-source console username
          password:           # Open-source console password
          registry:
            enabled: true     # Whether to enable service registration

4. Service Startup

@SpringBootApplication
public class Application {

    public static void main(String[] args) {
        SpringApplication.run(Application.class, args);
    }

    @Bean
    public ToolCallbackProvider weatherTools(WeatherService weatherService) {
        return MethodToolCallbackProvider.builder().toolObjects(weatherService).build();
    }

}

Developing an MCP Server Using Nacos MCP Wrapper Python

1. Environment Setup

pip install nacos-mcp-wrapper-python

2. Auto-registration Configuration

nacos_settings = NacosSettings()
nacos_settings.SERVER_ADDR = "127.0.0.1:8848" # <nacos_server_addr> e.g. 127.0.0.1:8848
nacos_settings.NAMESPACE= "public" # Nacos namespace ID
nacos_settings.USERNAME="" # Open-source console username
nacos_settings.PASSWORD="" # Open-source console password

3. Code Implementation

from nacos_mcp_wrapper.server.nacos_mcp import NacosMCP
from nacos_mcp_wrapper.server.nacos_settings import NacosSettings

# Create an MCP server instance
nacos_settings = NacosSettings()
nacos_settings.SERVER_ADDR = "127.0.0.1:8848" # <nacos_server_addr> e.g. 127.0.0.1:8848
nacos_settings.USERNAME=""
nacos_settings.PASSWORD=""
mcp = NacosMCP("nacos-mcp-python", nacos_settings=nacos_settings, port=18001)

# Register an addition tool
@mcp.tool()
def add(a: int, b: int) -> int:
    """Add two integers together"""
    return a + b

# Register a subtraction tool
@mcp.tool()
def minus(a: int, b: int) -> int:
    """Subtract two numbers"""
    return a - b


if __name__ == "__main__":
    try:
        mcp.run(transport="sse")
        # mcp.run(transport="stdio")
        # mcp.run(transport="streamable-http")
    except Exception as e:
        print(f"Runtime error: {e}")

For more usage examples, please refer to https://github.com/nacos-group/nacos-mcp-wrapper-python/tree/main

Dynamically Updating Tool Status and Descriptions in Nacos

After the MCP Server is automatically registered into Nacos, you can dynamically change the description and enable/disable status of the MCP Server Tools on the console. These changes will take effect at runtime on the MCP Server without requiring a service restart.

Frequently Asked Questions

Error mcp server info is not compatible or check mcp server compatible false occurs during startup

This issue occurs when there is already MCP Server data of the same version in Nacos, and the newly registered MCP Server data is incompatible with the existing one. Please ensure compatibility between the auto-registered MCP Server and the corresponding data in Nacos, including:

Consistent MCP Server protocol (stdio, sse, streamable)
Consistent service references
Consistent tool data (tool names, number of tools, parameter types, optional parameters). Note that tool descriptions and parameter descriptions are not required to match.

Discovering and Invoking the MCP Server

Discovering and Invoking the MCP Server Using the Spring AI Alibaba Nacos MCP Framework

1.Dependency Configuration

<dependency>
    <groupId>org.springframework.ai</groupId>
    <artifactId>spring-ai-autoconfigure-model-openai</artifactId>
    <version>{version >= 1.0.0}</version>
</dependency>

<dependency>
    <groupId>org.springframework.ai</groupId>
    <artifactId>spring-ai-autoconfigure-model-chat-client</artifactId>
    <version>{version >= 1.0.0}</version>
</dependency>

<dependency>
    <groupId>com.alibaba.cloud.ai</groupId>
    <artifactId>spring-ai-alibaba-starter-mcp-registry</artifactId>
    <version>{version >= 1.0.0.3}</version>
</dependency>

<dependency>
    <groupId>org.springframework.ai</groupId>
    <artifactId>spring-ai-starter-mcp-client-webflux</artifactId>
    <version>{version >= 1.0.0}</version>
</dependency>

2.Create Async or Sync MCP Client

    @Autowired
private List<LoadbalancedMcpSyncClient> loadbalancedMcpSyncClients;

    @Autowired
private List<LoadbalancedMcpAsyncClient> loadbalancedMcpAsyncClients;

Or Use ToolCallbackProvider

@Qualifier("loadbalancedSyncMcpToolCallbacks") ToolCallbackProvider tools

@Qualifier("loadbalancedMcpAsyncToolCallbacks") ToolCallbackProvider tools

3.application.yml配置

spring:
  application:
    name: mcp-client-webflux
  ai:
    openai:
      api-key: ${DASHSCOPE_API_KEY}
      base-url: https://dashscope.aliyuncs.com/compatible-mode
      chat:
        options:
          model: qwen-max
    alibaba:
      mcp:
        nacos:
          namespace: 4ad3108b-4d44-43d0-9634-3c1ac4850c8c
          server-addr: 127.0.0.1:8848
          username: nacos
          password: nacos
          client:
            enabled: true
            sse:
              connections:
                server1:
                  service-name: webflux-mcp-server # Nacos 中对应 MCP Server 的服务名
                  version: 1.0.0 # Nacos 中对应 MCP Server 的版本号
    mcp:
      client:
        enabled: true
        name: mcp-client-webflux
        version: 0.0.1
        initialized: true
        request-timeout: 600s
        type: sync
        toolcallback:
          enabled: true
        root-change-notification: true

4.Exmaple Code

@SpringBootApplication
public class NacosMcpDiscoveryClientApplication {


    public static void main(String[] args) {
        SpringApplication.run(NacosMcpDiscoveryClientApplication.class, args);
    }

    @Bean
    public CommandLineRunner predefinedQuestionsDemo(ChatClient.Builder chatClientBuilder, @Qualifier("loadbalancedMcpAsyncToolCallbacks") ToolCallbackProvider tools,
            ConfigurableApplicationContext context) {

        return args -> {
            var chatClient = chatClientBuilder
                    .defaultToolCallbacks(tools.getToolCallbacks())
                    .build();

            Scanner scanner = new Scanner(System.in);
            while (true) {
                System.out.print("\n>>> QUESTION: ");
                String userInput = scanner.nextLine();
                if (userInput.equalsIgnoreCase("exit")) {
                    break;
                }
                if (userInput.isEmpty()) {
                    userInput = "What the time in Beijing";
                }
                System.out.println("\n>>> ASSISTANT: " + chatClient.prompt(userInput).call().content());
            }
            scanner.close();
            context.close();
        };
    }
}

For the complete code, please refer to Spring AI Alibaba Examples - Nacos MCP Server Consumer Example

Discovering and Invoking MCP Server in Nacos Using Nacos MCP Router

Please refer to Nacos MCP Router

Discovering and Invoking MCP Server in Nacos within Dify

Please refer to Dify Discovers Nacos MCP Server