MCP server terminates if not kept alive explicitly via onCloseCallback

[SebastianAigner]

    runBlocking {
        server.connect(transport)
        val done = Job()
        server.onCloseCallback = {
            done.complete()
        }
        done.join()
    }

removing the done = Job() related logic from this causes the MCP server to terminate immediately, rather than wait for an process incoming requests.

[JsFlo]

Is this an issue ? you want the server to "not terminate forever unless you get the event in "onCloseCallback" and this seems like the right/idiomatic way to do that (job.join()).

A different way to do the above would be

runBlocking {
    server.connect(transport)
    val done = CompletableDeferred<Unit>()
    server.onCloseCallback = {
        done.complete(Unit)
    }
    done.await()
}

and if you really didn't want to handle the onCloseCallback (like it seems in your case since you removed the job) then you could just wait forever

runBlocking {
    server.connect(transport)
    awaitCancellation()
}