toad-eye 2.9.1 → 2.10.0

package/README.md

@@ -169,4 +169,4 @@ Full metrics and span attribute reference: [COMPATIBILITY.md](https://github.com
 
 ## Tech Stack
 
-TypeScript · OpenTelemetry SDK 2.x · Hono · Docker Compose (Prometheus, Jaeger, Grafana, OTel Collector) · Vitest (278+ tests)
+TypeScript · OpenTelemetry SDK 2.x · Hono · Docker Compose (Prometheus, Jaeger, Grafana, OTel Collector) · Vitest (285+ tests)

package/dist/alerts/channels.js

@@ -7,7 +7,7 @@ function smtpCacheKey(config) {
 }
 function formatMessage(alert) {
     const topModelsText = alert.topModels.length > 0
-        ? `\nTop models by spend:\n${alert.topModels.map((m) => `  • ${m.model}: $${m.value.toFixed(4)}`).join("\n")}`
+        ? `\nTop contributors:\n${alert.topModels.map((m) => `  • ${m.model}: ${m.value.toFixed(4)}`).join("\n")}`
         : "";
     return [
         `🚨 Alert: ${alert.rule.name}`,
@@ -27,6 +27,7 @@ async function sendTelegram(config, alert) {
             chat_id: config.chatId,
             text: formatMessage(alert),
         }),
+        signal: AbortSignal.timeout(10_000),
     });
     if (!res.ok)
         throw new Error(`Telegram send failed: ${res.status}`);
@@ -36,6 +37,7 @@ async function sendSlackWebhook(config, alert) {
         method: "POST",
         headers: { "Content-Type": "application/json" },
         body: JSON.stringify({ text: formatMessage(alert) }),
+        signal: AbortSignal.timeout(10_000),
     });
     if (!res.ok)
         throw new Error(`Slack webhook failed: ${res.status}`);
@@ -44,6 +46,7 @@ async function sendWebhook(config, alert) {
     const res = await fetch(config.url, {
         method: "POST",
         headers: { "Content-Type": "application/json", ...config.headers },
+        signal: AbortSignal.timeout(10_000),
         body: JSON.stringify({
             alertName: alert.rule.name,
             metric: alert.rule.metric,

package/dist/alerts/conditions.js

@@ -93,7 +93,7 @@ function matches(value, comparator, threshold) {
 }
 async function queryScalar(prometheusUrl, promql) {
     const url = `${prometheusUrl}/api/v1/query?query=${encodeURIComponent(promql)}`;
-    const res = await fetch(url);
+    const res = await fetch(url, { signal: AbortSignal.timeout(10_000) });
     if (!res.ok)
         throw new Error(`Prometheus query failed: ${res.status} ${res.statusText}`);
     const data = (await res.json());
@@ -103,7 +103,7 @@ async function queryScalar(prometheusUrl, promql) {
 async function queryTopModels(prometheusUrl, metric, window) {
     const promql = buildTopModelsQuery(metric, window);
     const url = `${prometheusUrl}/api/v1/query?query=${encodeURIComponent(promql)}`;
-    const res = await fetch(url);
+    const res = await fetch(url, { signal: AbortSignal.timeout(10_000) });
     if (!res.ok)
         return [];
     const labelKey = isMcpMetric(metric)

package/dist/alerts/grafana.js

@@ -12,6 +12,7 @@ export async function postGrafanaAnnotation(grafanaUrl, alert, apiKey) {
     const res = await fetch(`${grafanaUrl}/api/annotations`, {
         method: "POST",
         headers,
+        signal: AbortSignal.timeout(10_000),
         body: JSON.stringify({
             time: Date.now(),
             tags: ["toad-eye", "alert", alert.name],

package/dist/alerts/manager.js

@@ -14,6 +14,8 @@ export class AlertManager {
         this.config = config;
     }
     start() {
+        if (this.intervalId !== undefined)
+            return;
         const configured = this.config.evalIntervalSeconds ?? DEFAULT_EVAL_INTERVAL_SECONDS;
         const clampedSeconds = Math.max(configured, MIN_EVAL_INTERVAL_SECONDS);
         if (configured < MIN_EVAL_INTERVAL_SECONDS) {

package/dist/core/span-end-processor.js

@@ -34,7 +34,7 @@ function toSpanEndData(span) {
     return {
         traceId: ctx.traceId,
         spanId: ctx.spanId,
-        parentSpanId: span.parentSpanId || undefined,
+        parentSpanId: span.parentSpanContext?.spanId,
         name: span.name,
         kind: SPAN_KIND_MAP[span.kind] ?? "internal",
         status: span.status.code === SpanStatusCode.ERROR

package/dist/instrumentations/mcp.d.ts

@@ -8,5 +8,9 @@
  * Uses __toad_eye_mcp_patched flag to prevent double-instrumentation
  * when both auto and manual approaches are used.
  */
-export declare function enableMcpInstrumentation(): boolean;
+/**
+ * Enable MCP server auto-instrumentation.
+ * For ESM projects, pass McpServer class directly to avoid CJS/ESM module mismatch.
+ */
+export declare function enableMcpInstrumentation(McpServerClass?: any): boolean;
 export declare function disableMcpInstrumentation(): void;

package/dist/instrumentations/mcp.js

@@ -17,16 +17,25 @@ const MODULE_NAME = "@modelcontextprotocol/sdk/server/mcp.js";
 let mcpServerProto = null;
 // eslint-disable-next-line @typescript-eslint/no-explicit-any
 let originals = null;
-export function enableMcpInstrumentation() {
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+function resolveServerClass() {
     try {
-        require.resolve(MODULE_NAME);
+        // eslint-disable-next-line @typescript-eslint/no-require-imports
+        const sdk = require(MODULE_NAME);
+        return sdk.McpServer ?? sdk.default?.McpServer;
     }
     catch {
-        return false;
+        return null;
     }
-    // eslint-disable-next-line @typescript-eslint/no-require-imports
-    const sdk = require(MODULE_NAME);
-    const McpServer = sdk.McpServer ?? sdk.default?.McpServer;
+}
+/**
+ * Enable MCP server auto-instrumentation.
+ * For ESM projects, pass McpServer class directly to avoid CJS/ESM module mismatch.
+ */
+export function enableMcpInstrumentation(
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+McpServerClass) {
+    const McpServer = McpServerClass ?? resolveServerClass();
     if (!McpServer?.prototype)
         return false;
     if (McpServer.prototype[PATCHED_FLAG])

package/dist/mcp/client.d.ts

@@ -5,5 +5,9 @@
  * Injects W3C traceparent into _meta field so server-side middleware
  * can link client and server spans into one distributed trace.
  */
-export declare function enableMcpClientInstrumentation(): boolean;
+/**
+ * Enable MCP client instrumentation by auto-resolving the Client class.
+ * Works in CJS projects. For ESM projects, prefer passing the Client class directly.
+ */
+export declare function enableMcpClientInstrumentation(ClientClass?: any): boolean;
 export declare function disableMcpClientInstrumentation(): void;

package/dist/mcp/client.js

@@ -34,18 +34,35 @@ function injectTraceContext(params) {
     }
     return params;
 }
-export function enableMcpClientInstrumentation() {
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+function resolveClientClass() {
+    // Try CJS first (works when consumer uses require)
     try {
-        require.resolve("@modelcontextprotocol/sdk/client/index.js");
+        // eslint-disable-next-line @typescript-eslint/no-require-imports
+        const sdk = require("@modelcontextprotocol/sdk/client/index.js");
+        return sdk.Client ?? sdk.default?.Client;
     }
     catch {
-        return false;
+        // CJS failed — will try ESM in async path
+        return null;
     }
-    // eslint-disable-next-line @typescript-eslint/no-require-imports
-    const sdk = require("@modelcontextprotocol/sdk/client/index.js");
-    const Client = sdk.Client ?? sdk.default?.Client;
+}
+/**
+ * Enable MCP client instrumentation by auto-resolving the Client class.
+ * Works in CJS projects. For ESM projects, prefer passing the Client class directly.
+ */
+export function enableMcpClientInstrumentation(
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+ClientClass) {
+    const Client = ClientClass ?? resolveClientClass();
+    return Client ? patchClient(Client) : false;
+}
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+function patchClient(Client) {
     if (!Client?.prototype)
         return false;
+    if (typeof Client.prototype.callTool !== "function")
+        return false;
     if (Client.prototype[PATCHED_FLAG])
         return true;
     const proto = Client.prototype;

package/dist/mcp/middleware.js

@@ -33,6 +33,15 @@ function truncate(value, maxBytes) {
         return value;
     return value.slice(0, maxBytes) + "...[truncated]";
 }
+function redactValue(v, keys) {
+    if (Array.isArray(v)) {
+        return v.map((item) => redactValue(item, keys));
+    }
+    if (typeof v === "object" && v !== null) {
+        return redactObject(v, keys);
+    }
+    return v;
+}
 function redactObject(obj, keys) {
     if (typeof obj !== "object" || obj === null)
         return {};
@@ -41,11 +50,8 @@ function redactObject(obj, keys) {
         if (keys.includes(k)) {
             result[k] = "[REDACTED]";
         }
-        else if (typeof v === "object" && v !== null && !Array.isArray(v)) {
-            result[k] = redactObject(v, keys);
-        }
         else {
-            result[k] = v;
+            result[k] = redactValue(v, keys);
         }
     }
     return result;
@@ -74,7 +80,11 @@ function redactObject(obj, keys) {
  * In stdio MCP transport, stdout IS the JSON-RPC channel —
  * any stray console.log or OTel diagnostic would crash the connection.
  */
+let stdioSafe = false;
 function ensureStdioSafe() {
+    if (stdioSafe)
+        return;
+    stdioSafe = true;
     const stderrLogger = new DiagConsoleLogger();
     // Override the logger to use stderr-only methods
     const safeLogger = {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "toad-eye",
-  "version": "2.9.1",
+  "version": "2.10.0",
   "description": "Observability for MCP servers and LLM applications — traces, metrics, dashboards. One-line instrumentation for OpenAI, Anthropic, Gemini, and MCP. Self-hosted with Grafana + Prometheus + Jaeger.",
   "type": "module",
   "main": "dist/index.js",

package/templates/grafana/dashboards/agent-workflow.json

@@ -118,7 +118,7 @@
       },
       "targets": [
         {
-          "expr": "count(count by (gen_ai_agent_tool_name) (gen_ai_agent_tool_usage_total))",
+          "expr": "count(count by (gen_ai_tool_name) (gen_ai_agent_tool_usage_total))",
           "refId": "A"
         }
       ],
@@ -175,8 +175,8 @@
       },
       "targets": [
         {
-          "expr": "sort_desc(sum by (gen_ai_agent_tool_name) (gen_ai_agent_tool_usage_total))",
-          "legendFormat": "{{gen_ai_agent_tool_name}}",
+          "expr": "sort_desc(sum by (gen_ai_tool_name) (gen_ai_agent_tool_usage_total))",
+          "legendFormat": "{{gen_ai_tool_name}}",
           "refId": "A"
         }
       ]
@@ -231,8 +231,8 @@
       },
       "targets": [
         {
-          "expr": "sum by (gen_ai_agent_tool_name) (rate(gen_ai_agent_tool_usage_total[5m]))",
-          "legendFormat": "{{gen_ai_agent_tool_name}}",
+          "expr": "sum by (gen_ai_tool_name) (rate(gen_ai_agent_tool_usage_total[5m]))",
+          "legendFormat": "{{gen_ai_tool_name}}",
           "refId": "A"
         }
       ],
@@ -283,45 +283,13 @@
   "schemaVersion": 39,
   "tags": ["toad-eye", "agent"],
   "templating": {
-    "list": [
-      {
-        "name": "service",
-        "type": "query",
-        "datasource": {
-          "type": "prometheus",
-          "uid": "prometheus"
-        },
-        "query": "label_values(gen_ai_client_requests_total, job)",
-        "includeAll": true,
-        "allValue": ".*",
-        "current": {
-          "text": "All",
-          "value": "$__all"
-        },
-        "multi": true
-      },
-      {
-        "name": "tool",
-        "type": "query",
-        "datasource": {
-          "type": "prometheus",
-          "uid": "prometheus"
-        },
-        "query": "label_values(gen_ai_agent_tool_usage_total, gen_ai_agent_tool_name)",
-        "includeAll": true,
-        "multi": true,
-        "current": {
-          "text": "All",
-          "value": "$__all"
-        },
-        "regex": ""
-      }
-    ]
+    "list": []
   },
   "time": {
     "from": "now-6h",
     "to": "now"
   },
+  "refresh": "30s",
   "timepicker": {},
   "timezone": "utc",
   "title": "toad-eye / Agent Workflow",

package/templates/grafana/dashboards/finops-attribution.json

@@ -419,6 +419,7 @@
     "from": "now-7d",
     "to": "now"
   },
+  "refresh": "1m",
   "timepicker": {},
   "timezone": "utc",
   "title": "toad-eye / FinOps Attribution",

package/templates/grafana/dashboards/mcp-analytics.json

@@ -287,6 +287,7 @@
     "from": "now-1h",
     "to": "now"
   },
+  "refresh": "30s",
   "timepicker": {},
   "timezone": "browser",
   "title": "MCP Tool Analytics",

package/templates/grafana/dashboards/mcp-e2e.json

@@ -43,7 +43,7 @@
       "datasource": { "type": "prometheus", "uid": "prometheus" },
       "targets": [
         {
-          "expr": "sum(rate(gen_ai_client_operation_duration_sum[5m])) / sum(rate(gen_ai_client_operation_duration_count[5m])) or vector(0)",
+          "expr": "sum(rate(gen_ai_client_operation_duration_milliseconds_sum[5m])) / sum(rate(gen_ai_client_operation_duration_milliseconds_count[5m])) or vector(0)",
           "refId": "A"
         }
       ],
@@ -124,7 +124,7 @@
       "datasource": { "type": "prometheus", "uid": "prometheus" },
       "targets": [
         {
-          "expr": "sum(rate(gen_ai_client_operation_duration_sum[5m])) / sum(rate(gen_ai_client_operation_duration_count[5m]))",
+          "expr": "sum(rate(gen_ai_client_operation_duration_milliseconds_sum[5m])) / sum(rate(gen_ai_client_operation_duration_milliseconds_count[5m]))",
           "legendFormat": "LLM Avg Latency",
           "refId": "A"
         },
@@ -315,6 +315,7 @@
     "from": "now-1h",
     "to": "now"
   },
+  "refresh": "30s",
   "timepicker": {},
   "timezone": "browser",
   "title": "MCP End-to-End",

package/templates/grafana/dashboards/provider-health.json

@@ -166,7 +166,7 @@
       }
     },
     {
-      "title": "Error Breakdown by Type",
+      "title": "Error Breakdown by Model",
       "type": "timeseries",
       "gridPos": {
         "h": 8,
@@ -180,8 +180,8 @@
       },
       "targets": [
         {
-          "expr": "sum by (gen_ai_provider_name, error_type) (rate(gen_ai_client_errors_total{job=~\"$service\", gen_ai_provider_name=~\"$provider\"}[5m]))",
-          "legendFormat": "{{gen_ai_provider_name}} — {{error_type}}",
+          "expr": "sum by (gen_ai_provider_name, gen_ai_request_model) (rate(gen_ai_client_errors_total{job=~\"$service\", gen_ai_provider_name=~\"$provider\"}[5m]))",
+          "legendFormat": "{{gen_ai_provider_name}} — {{gen_ai_request_model}}",
           "refId": "A"
         }
       ],
@@ -276,6 +276,7 @@
         },
         "query": "label_values(gen_ai_client_requests_total, gen_ai_provider_name)",
         "includeAll": true,
+        "allValue": ".*",
         "multi": true,
         "current": {
           "text": "All",
@@ -289,6 +290,7 @@
     "from": "now-6h",
     "to": "now"
   },
+  "refresh": "30s",
   "timepicker": {},
   "timezone": "utc",
   "title": "toad-eye / Provider Health",