Spring AI How spring-ai-starter-mcp-server-webflux honours non-blocking behaviour with Tool response being always synchronous?

I'm exploring spring-ai-starter-mcp-server-webflux project to spin up MCP server and my expectation with this offering is the tool call processing should be asynchronous and non-blocking.

However, as per the documentation mentioned here on tool calling capabilities, the Mono or Flux is not supported. Which means,

Valid @Tool implementation:

@Tool(...)
public String sampleTool(..) {
return "test-string";
}

Invalid @Tool implementation:

@Tool(...)
public Mono<String> sampleTool(..) {
return Mono.just("test-string");
}

My question here is, if the implementation of @Tool is synchronous - then if I've to call any remote API / make data base calls, then my code bound to I/O and my threads will be blocked. Since I've to call .block() to return a String.

I'm trying to understand, where the concept of asynchrony / non-blocking here?

Comment From: YunKuiLu

Maybe you can try using Mcp Annotations in 1.1.0-SNAPSHOT.

@McpTool: https://docs.spring.io/spring-ai/reference/1.1-SNAPSHOT/api/mcp/mcp-annotations-server.html#_mcptool

Async Tool Server Example: https://docs.spring.io/spring-ai/reference/1.1-SNAPSHOT/api/mcp/mcp-annotations-examples.html#_async_tool_server

Comment From: deshpandeanoop

@YunKuiLu - So, is @Tool is blocking & should consider moving to @McpTool for async non-blocking behaviour. Just still wondering how does the current spring-ai-starter-mcp-server-webflux has to offer, since the whole implementation of the logic that goes under tool made sync.