Spring AI MCP server Error state [CLOSE_CONNECTION_NOW] reported while processing request (SSE+HTTP)

Bug description I'm using the following versions: spring-ai-starter-mcp-server-webmvc:1.0.0 (MPC server) spring-ai-starter-mcp-client:1.0.0 (MPC client) The client successfully sends requests and generates responses. However, most of the time, the client does not wait for the server's full response and instead returns prematurely with a default output. As a result,The connection between the client and server is aborted.

Server throws the following exception: java.io.EOFException: null at org.apache.tomcat.util.net.NioEndpoint$NioSocketWrapper.fillReadBuffer(NioEndpoint.java:1295) ~[tomcat-embed-core-10.1.41.jar:10.1.41] at org.apache.tomcat.util.net.NioEndpoint$NioSocketWrapper.read(NioEndpoint.java:1183) ~[tomcat-embed-core-10.1.41.jar:10.1.41]

025-08-03T12:15:41.002+10:00 DEBUG 20823 --- [sse-mcp-server-demo] [nio-8080-exec-2] o.apache.coyote.http11.Http11Processor : Error state [CLOSE_CONNECTION_NOW] reported while processing request

2025-08-03T12:15:41.002+10:00 DEBUG 20823 --- [sse-mcp-server-demo] [nio-8080-exec-2] o.apache.coyote.http11.Http11Processor : Socket: [org.apache.tomcat.util.net.NioEndpoint$NioSocketWrapper@7514622d:org.apache.tomcat.util.net.NioChannel@7bd392ff:java.nio.channels.SocketChannel[connected local=/[127.0.0.1:8080](http://127.0.0.1:8080/) remote=/127.0.0.1:65117]], Status in: [OPEN_READ], State out: [CLOSED]

2025-08-03T12:21:46.570+10:00 DEBUG 20823 --- [sse-mcp-server-demo] [ SIGINT handler] java.lang.Runtime : Runtime.exit() called with status: 130

However request header is not massive, printed in server logs connection: Upgrade, HTTP2-Settings content-length: 69 host: localhost:8080 http2-settings: AAEAAEAAAAIAAAAAAAMAAAAAAAQBAAAAAAUAAEAAAAYABgAA upgrade: h2c user-agent: Java-http-client/21.0.8 accept: text/event-stream cache-control: no-cache content-type: application/json

Environment Spring AI version: 1.0.0 Java version: 21 Running in local host

Steps to reproduce

1. Start the server.
2. Start the client.
3. Observe the logs — they indicate that the client and server are communicating initially.
4. The server responds to one or two messages, then throws the error mentioned above.
5. The client continues running but returns a default response, stating that no tool is defined. Client responds before server returns with 2025-08-03T12:15:18.586+10:00 INFO 20994 --- [nio-8081-exec-1] mcp_client.ClientController : User message: get all stores 2025-08-03T12:15:18.586+10:00 INFO 20994 --- [nio-8081-exec-1] mcp_client.ClientController : AI response: To get all stores, you can use the following function: python stores = spring_ai_mcp_client_server1_getAllStores()

Expected behavior The client should wait for the server's response and return that response, rather than falling back to a default message.

Minimal Complete Reproducible example This is server application.yaml spring: application: name: sse-mcp-server-demo ai: mcp: server: enabled: true stdio: false name: sse-mcp-server-demo version: 1.0.0 resource-change-notification: true tool-change-notification: true prompt-change-notification: true sse-endpoint: /api/v1/sse sse-message-endpoint: /api/v1/mcp type: sync capabilities: completion: true prompt: true resource: true tool: true server: error: include-message: always max-http-request-header-size: 640

This is client application.yaml server: port: 8081 spring: application: ollama: base-url: http://localhost:11434/ timeout: 300s chat: enabled: true options: model: qwen3:8b ai: mcp: client: enabled: true name: spring-ai-mcp-client version: 1.0.0 initialized: true request-timeout: 300s type: sync root-change-notification: true toolcallback.enabled: true sse: connections: server1: url: http://localhost:8080/ sse-endpoint: /api/v1/sse

Client is using chat client @Bean fun chatClient(builder: ChatClient.Builder, toolCallbackProvider: ToolCallbackProvider): ChatClient { return builder .defaultToolCallbacks(toolCallbackProvider) .defaultAdvisors( MessageChatMemoryAdvisor.builder( MessageWindowChatMemory.builder().build() ) .build() ) .build(); }

Service using chat client to prompt @Service class AiServiceImpl(val chatClient: ChatClient) : AiService { override fun complete(message: String): String? { return chatClient.prompt() .user(message) .call() .content(); } }

Comment From: wilocu

I tried to recreate the issue and found that:

Spring AI Schema Generation Spring AI appears to generate correct schemas for parameter-less functions: {"type": "object", "properties": {}, "required": []}

The issue might be in the external MCP SDK's WebFlux transport (WebClientStreamableHttpTransport) rather than Spring AI itself.

Comment From: wathsala

Thanks for looking into this @wilocu , I am using webmvc, not webflux

Comment From: wilocu

I was stuck so I used AI to try and find the issue, not sure if this is correct but it does look plausible: Since you're using spring-ai-starter-mcp-server-webmvc, your client likely uses HttpClientSseClientTransport. The premature response/"CLOSE_CONNECTION_NOW" issue appears to be in the external MCP SDK's HttpClientSseClientTransport - not Spring AI's
code.

Potential solutions: 1. Try switching to WebFlux client transport 2. Check timeout configurations 3. Report to external MCP SDK repo

Comment From: wathsala

Thanks @wilocu , I have tried with webflux Regarding WebFlux implementation, can we write tools without blocking? Below is ToolCallbackProvider implementation. Is there a separate implementation for async mode? Because with this MCP inspector and with MCP client server does not respond if I do not block the tool implementation.

@Bean open fun bindTools(searchTools: ProductSearchTools): ToolCallbackProvider { return MethodToolCallbackProvider.builder() .toolObjects(searchTools, recipesTools, deliveryTrackerTools, slotsTools) .build() }

Server logs when return Mono response from Tool implementation

--- [ai-agent-mcp-sample-java] [ctor-http-nio-6] i.m.spec.McpServerSession                : Received request: JSONRPCRequest[jsonrpc=2.0, method=tools/call, id=4, params={_meta={progressToken=4}, name=getProductSuggestionsResults, arguments={query=milk}}] --- [ai-agent-mcp-sample-java] [oundedElastic-1] o.s.ai.tool.method.MethodToolCallback    : Starting execution of tool: getProductSuggestionsResults --- [ai-agent-mcp-sample-java] [oundedElastic-1] o.s.ai.tool.method.MethodToolCallback    : Successful execution of tool: getProductSuggestionsResults --- [ai-agent-mcp-sample-java] [oundedElastic-1] o.s.a.t.e.DefaultToolCallResultConverter : Converting tool result to JSON. --- [ai-agent-mcp-sample-java] [oundedElastic-1] o.s.w.s.adapter.HttpWebHandlerAdapter    : [dc65e3b7-7] Completed 200 OK = --- [ai-agent-mcp-sample-java] [ctor-http-nio-6] r.netty.channel.ChannelOperations        : [dc65e3b7-1, L:/[127.0.0.1:8080](http://127.0.0.1:8080/) - R:/[127.0.0.1:49972](http://127.0.0.1:49972/)] [HttpServer] Channel inbound receiver cancelled (subscription disposed).

MCP Inspector