Fix vMCP metric naming to align with Prometheus conventions

docs/operator/virtualmcpserver-observability.md

@@ -39,7 +39,7 @@ the workflow name as an attribute.
 
 ## Distributed Tracing
 
-The vMCP creates spans for each individual backend operation as well as workflow executions, enabling the attribution of workflow exection errors or latency to specific tool calls.
+The vMCP creates spans for each individual backend operation as well as workflow executions, enabling the attribution of workflow execution errors or latency to specific tool calls.
 
 
 ## Configuration

pkg/vmcp/server/server.go

@@ -204,7 +204,7 @@
 	// backend calls are instrumented when they occur during workflow execution.
 	if cfg.TelemetryProvider != nil {
 		var err error
 		backendClient, err = monitorBackends(context.Background(), cfg.TelemetryProvider.MeterProvider(), cfg.TelemetryProvider.TracerProvider(), backends, backendClient)
 		if err != nil {
 			return nil, fmt.Errorf("failed to monitor backends: %w", err)
 		}
@@ -224,7 +224,7 @@
 
 	// Decorate workflow executors with telemetry if provider is configured
 	if cfg.TelemetryProvider != nil {
 		workflowExecutors, err = monitorWorkflowExecutors(cfg.TelemetryProvider.MeterProvider(), cfg.TelemetryProvider.TracerProvider(), workflowExecutors)
 		if err != nil {
 			return nil, fmt.Errorf("failed to monitor workflow executors: %w", err)
 		}
@@ -375,7 +375,7 @@
 		logger.Info("RFC 9728 OAuth discovery endpoints enabled at /.well-known/")
 	}
 
-	// MCP endpoint - apply middleware chain: auth → discovery
+	// MCP endpoint - apply middleware chain: auth → discovery → telemetry
 	var mcpHandler http.Handler = streamableServer
 
 	if s.config.TelemetryProvider != nil {

pkg/vmcp/server/telemetry.go

@@ -38,15 +38,19 @@
 	}
 	backendCount.Record(ctx, int64(len(backends)))
 
-	requestsTotal, err := meter.Int64Counter("toolhive_vmcp_backend_requests_total", metric.WithDescription("Total number of requests per backend"))
+	requestsTotal, err := meter.Int64Counter("toolhive_vmcp_backend_requests", metric.WithDescription("Total number of requests per backend"))
 	if err != nil {
 		return nil, fmt.Errorf("failed to create requests total counter: %w", err)
 	}
-	errorsTotal, err := meter.Int64Counter("toolhive_vmcp_backend_errors_total", metric.WithDescription("Total number of errors per backend"))
+	errorsTotal, err := meter.Int64Counter("toolhive_vmcp_backend_errors", metric.WithDescription("Total number of errors per backend"))
 	if err != nil {
 		return nil, fmt.Errorf("failed to create errors total counter: %w", err)
 	}
-	requestsDuration, err := meter.Float64Histogram("toolhive_vmcp_backend_requests_duration", metric.WithDescription("Duration of requests in seconds per backend"))
+	requestsDuration, err := meter.Float64Histogram(
+		"toolhive_vmcp_backend_requests_duration",
+		metric.WithDescription("Duration of requests in seconds per backend"),
+		metric.WithUnit("s"),
+	)
 	if err != nil {
 		return nil, fmt.Errorf("failed to create requests duration histogram: %w", err)
 	}
@@ -73,7 +77,7 @@
 
 // record updates the metrics and creates a span for each method on the BackendClient interface.
 // It returns a function that should be deferred to record the duration, error, and end the span.
 func (t telemetryBackendClient) record(ctx context.Context, target *vmcp.BackendTarget, action string, err *error) (context.Context, func()) {
 	// Create span attributes
 	commonAttrs := []attribute.KeyValue{
 		attribute.String("target.workload_id", target.WorkloadID),
@@ -104,25 +108,25 @@
 	}
 }
 
 func (t telemetryBackendClient) CallTool(ctx context.Context, target *vmcp.BackendTarget, toolName string, arguments map[string]any) (_ map[string]any, retErr error) {
 	ctx, done := t.record(ctx, target, "call_tool", &retErr)
 	defer done()
 	return t.backendClient.CallTool(ctx, target, toolName, arguments)
 }
 
 func (t telemetryBackendClient) ReadResource(ctx context.Context, target *vmcp.BackendTarget, uri string) (_ []byte, retErr error) {
 	ctx, done := t.record(ctx, target, "read_resource", &retErr)
 	defer done()
 	return t.backendClient.ReadResource(ctx, target, uri)
 }
 
 func (t telemetryBackendClient) GetPrompt(ctx context.Context, target *vmcp.BackendTarget, name string, arguments map[string]any) (_ string, retErr error) {
 	ctx, done := t.record(ctx, target, "get_prompt", &retErr)
 	defer done()
 	return t.backendClient.GetPrompt(ctx, target, name, arguments)
 }
 
 func (t telemetryBackendClient) ListCapabilities(ctx context.Context, target *vmcp.BackendTarget) (_ *vmcp.CapabilityList, retErr error) {
 	ctx, done := t.record(ctx, target, "list_capabilities", &retErr)
 	defer done()
 	return t.backendClient.ListCapabilities(ctx, target)
@@ -142,24 +146,25 @@
 	meter := meterProvider.Meter(instrumentationName)
 
 	executionsTotal, err := meter.Int64Counter(
-		"toolhive_vmcp_workflow_executions_total",
+		"toolhive_vmcp_workflow_executions",
 		metric.WithDescription("Total number of workflow executions"),
 	)
 	if err != nil {
 		return nil, fmt.Errorf("failed to create workflow executions counter: %w", err)
 	}
 
 	errorsTotal, err := meter.Int64Counter(
-		"toolhive_vmcp_workflow_errors_total",
+		"toolhive_vmcp_workflow_errors",
 		metric.WithDescription("Total number of workflow execution errors"),
 	)
 	if err != nil {
 		return nil, fmt.Errorf("failed to create workflow errors counter: %w", err)
 	}
 
 	executionDuration, err := meter.Float64Histogram(
-		"toolhive_vmcp_workflow_duration_seconds",
+		"toolhive_vmcp_workflow_duration",
 		metric.WithDescription("Duration of workflow executions in seconds"),
+		metric.WithUnit("s"),
 	)
 	if err != nil {
 		return nil, fmt.Errorf("failed to create workflow duration histogram: %w", err)