fix: send SIGHUP to MCP container after initial deploy

server/internal/orchestrator/swarm/mcp_config_resource.go

@@ -254,6 +254,21 @@ func (r *MCPConfigResource) writeUserFileIfNeeded(fs afero.Fs, usersPath string)
 // return nil — the config file is already correct on disk and will be
 // picked up on the next container restart. Injector failures are returned
 // as errors since they indicate a systemic problem.
+// PostDeploy is called by ServiceInstanceResource after the container is confirmed
+// running. It sends SIGHUP to trigger a config reload, handling VirtioFS
+// bind-mount propagation delays on initial provisioning.
+func (r *MCPConfigResource) PostDeploy(ctx context.Context, rc *resource.Context) {
+	logger, err := do.Invoke[zerolog.Logger](rc.Injector)
+	if err != nil {
+		return
+	}
+	if err := r.signalConfigReload(ctx, rc); err != nil {
+		logger.Warn().Err(err).
+			Str("service_instance_id", r.ServiceInstanceID).
+			Msg("post-deploy config reload signal failed")
+	}
+}
+
 func (r *MCPConfigResource) signalConfigReload(ctx context.Context, rc *resource.Context) error {
 	dockerClient, err := do.Invoke[*docker.Docker](rc.Injector)
 	if err != nil {

server/internal/orchestrator/swarm/service_instance.go

@@ -203,6 +203,8 @@ func (s *ServiceInstanceResource) deploy(ctx context.Context, rc *resource.Conte
 		Str("service_id", res.ServiceID).
 		Msg("service instance started successfully")
 
+	spec.PostDeploy(ctx, rc)
+
 	// Transition state to "running" immediately after successful deployment.
 	// This mirrors the database instance pattern where state is set
 	// deterministically within the resource activity, not deferred to the monitor.

server/internal/orchestrator/swarm/service_instance_spec.go

@@ -90,6 +90,22 @@ func (s *ServiceInstanceSpecResource) TypeDependencies() []resource.Type {
 	return nil
 }
 
+// PostDeploy is called by ServiceInstanceResource after the container is confirmed
+// running. Each service type can trigger any necessary post-start actions here.
+func (s *ServiceInstanceSpecResource) PostDeploy(ctx context.Context, rc *resource.Context) {
+	switch s.ServiceSpec.ServiceType {
+	case "mcp":
+		mcpCfg, err := resource.FromContext[*MCPConfigResource](rc, MCPConfigResourceIdentifier(s.ServiceInstanceID))
+		if err != nil {
+			log.Warn().Err(err).
+				Str("service_instance_id", s.ServiceInstanceID).
+				Msg("skipping MCP post-deploy hook: MCP config resource unavailable")
+			return
+		}
+		mcpCfg.PostDeploy(ctx, rc)
+	}
+}
+
 func (s *ServiceInstanceSpecResource) Refresh(ctx context.Context, rc *resource.Context) error {
 	network, err := resource.FromContext[*Network](rc, NetworkResourceIdentifier(s.DatabaseNetworkID))
 	if err != nil {