Make the server resilient to restart in SSE mode

[sebastienblanc]

Hi !
Not sure if my questions here or on the mcp-client side.
Basically I would like the mcp-server to be resilient to restarts, now of course after a restart, the SSE connection is lost and when a request comes in, I get something like :

2025-02-09 16:46:27,365 ERROR [io.qua.mcp.ser.sse.run.SseMcpMessageHandler] (vert.x-eventloop-thread-1) Connection not found: NmRkZTk4ZDQtNmY3My00NDQ0LWJjOWEtNTBkNTg4MDlmZTQ4

Is there a way to solve to that ? Like force a reconnection if the connection is not found ?

[sebastienblanc]

Maybe we could have https://github.com/quarkiverse/quarkus-mcp-server/blob/main/core/runtime/src/main/java/io/quarkiverse/mcp/server/runtime/ConnectionManager.java as an interface, so we could have a implementation that uses Redis for instance, to keep the connection details ?

[mkouba]

Hi Sebastien, that is a very good question. However, I'm not so sure there's something we can do on the server side. If an SSE connection is lost then the server has no way to send messages to the client and reconnection (and MCP reinitialization) is probably the only way to fix that.

Maybe we could have https://github.com/quarkiverse/quarkus-mcp-server/blob/main/core/runtime/src/main/java/io/quarkiverse/mcp/server/runtime/ConnectionManager.java as an interface, so we could have a implementation that uses Redis for instance, to keep the connection details ?

This would help to keep the session state but the SSE connection is lost anyway.

WDYT @jmartisk

[sebastienblanc]

Make sense. If I'm using quarkus-langchain4j-mcp on the client side, is there a way to catch the 400 response and initiate a new connection ?

[mkouba]

Make sense. If I'm using quarkus-langchain4j-mcp on the client side, is there a way to catch the 400 response and initiate a new connection ?

I have no idea, that's a question for Jan ;-).

[jmartisk]

Yeah, it's on my radar to see what we can do on the client side, thanks for opening the issue there. I assume it could try reconnecting automatically and just doing the whole init phase again, failing the currently active operations.
I assume this server-side issue can be closed now?

[mkouba]

I assume this server-side issue can be closed now?

Yes, I think so. FYI I've implemented the dummy initialization in the dev mode for SSE which basically means that the server is able to accept new requests after the hot reload if the client reconnects SSE.