diff --git a/pkg/epp/handlers/server.go b/pkg/epp/handlers/server.go
index 0fb7f7ba9..6e5fb2f0a 100644
--- a/pkg/epp/handlers/server.go
+++ b/pkg/epp/handlers/server.go
@@ -83,6 +83,7 @@ type RequestContext struct {
 	Usage                     Usage
 	ResponseSize              int
 	ResponseComplete          bool
+	FirstTokenReceived        bool
 	ResponseStatusCode        string
 	RequestRunning            bool
 	Request                   *Request
diff --git a/pkg/epp/requestcontrol/director.go b/pkg/epp/requestcontrol/director.go
index 844f0523b..69aa6ce57 100644
--- a/pkg/epp/requestcontrol/director.go
+++ b/pkg/epp/requestcontrol/director.go
@@ -298,10 +298,17 @@ func (d *Director) HandleResponseReceived(ctx context.Context, reqCtx *handlers.
 func (d *Director) HandleResponseBodyStreaming(ctx context.Context, reqCtx *handlers.RequestContext) (*handlers.RequestContext, error) {
 	logger := log.FromContext(ctx).WithValues("stage", "bodyChunk")
 	logger.V(logutil.TRACE).Info("Entering HandleResponseBodyChunk")
+
+	isFirstToken := !reqCtx.FirstTokenReceived
+	if isFirstToken {
+		reqCtx.FirstTokenReceived = true
+	}
+
 	response := &Response{
-		RequestId:   reqCtx.Request.Headers[requtil.RequestIdHeaderKey],
-		Headers:     reqCtx.Response.Headers,
-		EndOfStream: reqCtx.ResponseComplete,
+		RequestId:    reqCtx.Request.Headers[requtil.RequestIdHeaderKey],
+		Headers:      reqCtx.Response.Headers,
+		IsFirstToken: isFirstToken,
+		EndOfStream:  reqCtx.ResponseComplete,
 	}
 
 	d.runResponseStreamingPlugins(ctx, reqCtx.SchedulingRequest, response, reqCtx.TargetPod)
diff --git a/pkg/epp/requestcontrol/types.go b/pkg/epp/requestcontrol/types.go
index 7a6725678..4cc50034b 100644
--- a/pkg/epp/requestcontrol/types.go
+++ b/pkg/epp/requestcontrol/types.go
@@ -26,6 +26,10 @@ type Response struct {
 	Body string
 	// IsStreaming indicates whether or not the response is being streamed by the model
 	IsStreaming bool
+	// IsFirstToken when true indicates this is the first chunk of a streaming response.
+	// This is useful for plugins that need to perform actions at the time-to-first-token (TTFT) moment,
+	// such as marking prefill completion in disaggregated inference architectures.
+	IsFirstToken bool
 	// EndOfStream when true indicates that this invocation contains the last chunk of the response
 	EndOfStream bool
 	// ReqMetadata is a map of metadata that can be passed from Envoy.