toad-eye 2.6.0 → 2.7.0

package/README.md

@@ -46,13 +46,14 @@ Open [localhost:3100](http://localhost:3100) (Grafana, admin/admin) to see your
 
 ## What you get
 
-| Feature                  | Description                                                                                |
-| ------------------------ | ------------------------------------------------------------------------------------------ |
-| **Auto-instrumentation** | OpenAI, Anthropic, Gemini, Vercel AI SDK — regular + streaming                             |
-| **8 Grafana dashboards** | Overview, Cost, Latency, Errors, Model Comparison, FinOps, Provider Health, Agent Workflow |
-| **Cost tracking**        | Per-request USD cost, daily totals, projected monthly spend                                |
-| **Budget guards**        | Daily/per-user/per-model spend limits — warn, block, or auto-downgrade                     |
-| **Privacy controls**     | Built-in PII redaction (email, SSN, CC, phone), hashing, content masking                   |
+| Feature                   | Description                                                                            |
+| ------------------------- | -------------------------------------------------------------------------------------- |
+| **Auto-instrumentation**  | OpenAI, Anthropic, Gemini, Vercel AI SDK — regular + streaming                         |
+| **MCP server monitoring** | One-line middleware for MCP servers — trace every tool call, resource read, prompt     |
+| **9 Grafana dashboards**  | Overview, Cost, Latency, Errors, Model Comparison, FinOps, Provider Health, Agent, MCP |
+| **Cost tracking**         | Per-request USD cost, daily totals, projected monthly spend                            |
+| **Budget guards**         | Daily/per-user/per-model spend limits — warn, block, or auto-downgrade                 |
+| **Privacy controls**      | Built-in PII redaction (email, SSN, CC, phone), hashing, content masking               |
 
 <details>
 <summary>Advanced features</summary>
@@ -124,38 +125,42 @@ const result = await traceLLMCall(
 
 </details>
 
-<details>
-<summary>MCP Server observability</summary>
+## MCP Server observability
 
-Drop-in middleware for MCP servers — one line for full tracing of every tool call, resource read, and prompt request.
+8,600+ MCP servers. Zero built-in observability. toad-eye fixes that with one line:
 
 ```typescript
-import { McpServer } from "@modelcontextprotocol/sdk/server/index.js";
+import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
+import { initObservability } from "toad-eye";
 import { toadEyeMiddleware } from "toad-eye/mcp";
 
+initObservability({ serviceName: "my-mcp-server" });
+
 const server = new McpServer({ name: "my-server", version: "1.0.0" });
-toadEyeMiddleware(server);
+toadEyeMiddleware(server); // every tool call, resource read, prompt — traced
 
-// All handlers are now instrumented — spans appear in Jaeger
 server.tool("calculate", { expression: z.string() }, async ({ expression }) => {
   return { content: [{ type: "text", text: String(eval(expression)) }] };
 });
 ```
 
-Spans follow OTel GenAI semconv: `execute_tool calculate`, `retrieval file:///docs`, `prompt greeting`.
+Spans follow [OTel MCP semantic conventions](https://opentelemetry.io/docs/specs/semconv/gen-ai/): `tools/call calculate`, `resources/read file:///docs`, `prompts/get greeting`.
+
+Every span includes: `mcp.method.name`, `mcp.session.id`, `mcp.protocol.version`, `network.transport`.
 
 Privacy by default — tool arguments and results are NOT recorded unless you opt in:
 
 ```typescript
 toadEyeMiddleware(server, {
   recordInputs: true,
+  recordOutputs: true,
   redactKeys: ["apiKey", "token"],
 });
 ```
 
-> Safe for stdio transport — OTel diagnostics are redirected to stderr.
+Dedicated MCP dashboard in Grafana: tool call rate, duration p50/p95, error rate, resource reads, tool performance table.
 
-</details>
+> Safe for stdio transport — OTel diagnostics are redirected to stderr.
 
 ## Budget guards
 
@@ -378,7 +383,7 @@ Self-hosted mode remains the default. Cloud mode activates automatically when `a
 
 ## Grafana dashboards
 
-8 pre-built dashboards auto-provisioned on `npx toad-eye init`:
+9 pre-built dashboards auto-provisioned on `npx toad-eye init`:
 
 | Dashboard              | What it shows                                                    |
 | ---------------------- | ---------------------------------------------------------------- |
@@ -390,6 +395,7 @@ Self-hosted mode remains the default. Cloud mode activates automatically when `a
 | **FinOps Attribution** | Cost by team/user/feature, projected spend, what-if table        |
 | **Provider Health**    | Provider status (healthy/degraded/down), uptime, error breakdown |
 | **Agent Workflow**     | Steps per query, tool usage frequency, step type breakdown       |
+| **MCP Server**         | Tool call rate, duration p50/p95, errors by tool, resource reads |
 
 ## Subpath imports
 
@@ -448,24 +454,24 @@ import { ToadEyeAISpanProcessor, withToadEye } from "toad-eye/vercel"; // Vercel
 
 ### Span attributes (OTel GenAI semconv)
 
-| Attribute                          | Description                                     |
-| ---------------------------------- | ----------------------------------------------- |
-| `gen_ai.operation.name`            | `chat`, `invoke_agent`, `execute_tool`          |
-| `gen_ai.provider.name`             | `anthropic`, `gemini`, `openai`                 |
-| `gen_ai.request.model`             | Model identifier                                |
-| `gen_ai.usage.input_tokens`        | Tokens in the prompt                            |
-| `gen_ai.usage.output_tokens`       | Tokens in the completion                        |
-| `gen_ai.request.temperature`       | Temperature parameter                           |
-| `gen_ai.agent.name`                | Agent name (for agent spans)                    |
-| `gen_ai.agent.id`                  | Agent identifier                                |
-| `gen_ai.tool.name`                 | Tool name for execute_tool spans                |
-| `gen_ai.tool.type`                 | `function`, `extension`, `retrieval`, `builtin` |
-| `gen_ai.toad_eye.cost`             | Cost in USD                                     |
-| `gen_ai.toad_eye.agent.step.type`  | ReAct step: think/act/observe/answer/handoff    |
-| `gen_ai.toad_eye.agent.handoff.to` | Target agent for handoff steps                  |
-| `gen_ai.toad_eye.agent.loop_count` | Agent loop iterations (observe→think)           |
-| `gen_ai.toad_eye.guard.mode`       | Guard mode: shadow/enforce                      |
-| `session.id`                       | Session identifier (if configured)              |
+| Attribute                          | Description                                                            |
+| ---------------------------------- | ---------------------------------------------------------------------- |
+| `gen_ai.operation.name`            | `chat`, `invoke_agent`, `execute_tool`, `tools/call`, `resources/read` |
+| `gen_ai.provider.name`             | `anthropic`, `gemini`, `openai`                                        |
+| `gen_ai.request.model`             | Model identifier                                                       |
+| `gen_ai.usage.input_tokens`        | Tokens in the prompt                                                   |
+| `gen_ai.usage.output_tokens`       | Tokens in the completion                                               |
+| `gen_ai.request.temperature`       | Temperature parameter                                                  |
+| `gen_ai.agent.name`                | Agent name (for agent spans)                                           |
+| `gen_ai.agent.id`                  | Agent identifier                                                       |
+| `gen_ai.tool.name`                 | Tool name for execute_tool spans                                       |
+| `gen_ai.tool.type`                 | `function`, `extension`, `retrieval`, `builtin`                        |
+| `gen_ai.toad_eye.cost`             | Cost in USD                                                            |
+| `gen_ai.toad_eye.agent.step.type`  | ReAct step: think/act/observe/answer/handoff                           |
+| `gen_ai.toad_eye.agent.handoff.to` | Target agent for handoff steps                                         |
+| `gen_ai.toad_eye.agent.loop_count` | Agent loop iterations (observe→think)                                  |
+| `gen_ai.toad_eye.guard.mode`       | Guard mode: shadow/enforce                                             |
+| `session.id`                       | Session identifier (if configured)                                     |
 
 </details>
 

package/dist/cli.js

@@ -97,19 +97,26 @@ function down() {
 export function findContainerByService(containers, serviceKey) {
     return containers.find((c) => c.Service === serviceKey);
 }
+function isContainerHealthy(container) {
+    return container?.State === "running" && container.ExitCode === 0;
+}
 function status() {
     const composeFile = requireInfra();
     try {
-        const output = execFileSync("docker", ["compose", "-f", composeFile, "ps", "--format", "json"], { encoding: "utf-8" });
+        const output = execFileSync("docker", ["compose", "-f", composeFile, "ps", "--all", "--format", "json"], { encoding: "utf-8" });
         const lines = output.trim().split("\n").filter(Boolean);
         const containers = lines.map((line) => JSON.parse(line));
         console.log("\u{1f438} toad-eye stack:");
         console.log();
         for (const svc of SERVICES) {
-            const container = findContainerByService(containers, svc.service);
-            const state = container?.State === "running" ? "\u{1f7e2}" : "\u{1f534}";
+            const container = containers.find((c) => c.Service === svc.service);
+            const healthy = isContainerHealthy(container);
+            const icon = healthy ? "\u{1f7e2}" : "\u{1f534}";
+            const suffix = container && !healthy
+                ? ` (${container.State}${container.ExitCode ? `, exit code ${container.ExitCode}` : ""})`
+                : "";
             const loginInfo = "login" in svc ? ` (${svc.login})` : "";
-            console.log(`  ${state} ${svc.name.padEnd(16)} ${svc.url}${loginInfo}`);
+            console.log(`  ${icon} ${svc.name.padEnd(16)} ${svc.url}${loginInfo}${suffix}`);
         }
         console.log();
     }

package/dist/mcp/index.d.ts

@@ -1,3 +1,4 @@
 export { toadEyeMiddleware, traceSampling } from "./middleware.js";
 export type { TraceSamplingOptions } from "./middleware.js";
 export type { ToadMcpOptions, McpSpanAttributes } from "./types.js";
+export { MCP_METHODS } from "./spans.js";

package/dist/mcp/index.js

@@ -1 +1,2 @@
 export { toadEyeMiddleware, traceSampling } from "./middleware.js";
+export { MCP_METHODS } from "./spans.js";

package/dist/mcp/metrics.d.ts

@@ -3,6 +3,8 @@
  *
  * Separate from core metrics (core/metrics.ts) — these only exist
  * when MCP middleware is active. Initialized on first middleware call.
+ *
+ * All metrics include `mcp.method.name` dimension per OTel MCP semconv.
  */
 export declare function recordMcpToolCall(toolName: string, durationMs: number, status: "success" | "error"): void;
 export declare function recordMcpToolError(toolName: string, errorType: string): void;

package/dist/mcp/metrics.js

@@ -3,9 +3,12 @@
  *
  * Separate from core metrics (core/metrics.ts) — these only exist
  * when MCP middleware is active. Initialized on first middleware call.
+ *
+ * All metrics include `mcp.method.name` dimension per OTel MCP semconv.
  */
 import { metrics, } from "@opentelemetry/api";
 import { GEN_AI_METRICS } from "../types/metrics.js";
+import { MCP_METHODS } from "./spans.js";
 const METER_NAME = "toad-eye-mcp";
 let toolDuration;
 let toolCalls;
@@ -37,16 +40,30 @@ function ensureInit() {
 }
 export function recordMcpToolCall(toolName, durationMs, status) {
     ensureInit();
-    toolCalls.add(1, { "gen_ai.tool.name": toolName, status });
-    toolDuration.record(durationMs, { "gen_ai.tool.name": toolName });
+    toolCalls.add(1, {
+        "gen_ai.tool.name": toolName,
+        "mcp.method.name": MCP_METHODS.TOOLS_CALL,
+        status,
+    });
+    toolDuration.record(durationMs, {
+        "gen_ai.tool.name": toolName,
+        "mcp.method.name": MCP_METHODS.TOOLS_CALL,
+    });
 }
 export function recordMcpToolError(toolName, errorType) {
     ensureInit();
-    toolErrors.add(1, { "gen_ai.tool.name": toolName, "error.type": errorType });
+    toolErrors.add(1, {
+        "gen_ai.tool.name": toolName,
+        "mcp.method.name": MCP_METHODS.TOOLS_CALL,
+        "error.type": errorType,
+    });
 }
 export function recordMcpResourceRead(uri) {
     ensureInit();
-    resourceReads.add(1, { "gen_ai.data_source.id": uri });
+    resourceReads.add(1, {
+        "gen_ai.data_source.id": uri,
+        "mcp.method.name": MCP_METHODS.RESOURCES_READ,
+    });
 }
 export function recordMcpSessionStart() {
     ensureInit();

package/dist/mcp/middleware.d.ts

@@ -45,4 +45,10 @@ export interface TraceSamplingOptions {
     readonly serverVersion?: string | undefined;
     /** Parent OTel context. */
     readonly parentContext?: import("@opentelemetry/api").Context | undefined;
+    /** MCP session ID. */
+    readonly sessionId?: string | undefined;
+    /** MCP protocol version. */
+    readonly protocolVersion?: string | undefined;
+    /** Network transport: "pipe" (stdio) or "tcp" (HTTP). */
+    readonly networkTransport?: string | undefined;
 }

package/dist/mcp/middleware.js

@@ -9,10 +9,25 @@
  * and strict TypeScript. Wrapping the public API is the stable path.
  */
 import { context, diag, DiagConsoleLogger, DiagLogLevel, } from "@opentelemetry/api";
+import { randomUUID } from "node:crypto";
 import { extractContextFromMeta } from "./context.js";
 import { startToolSpan, startResourceSpan, startPromptSpan, startSamplingSpan, endSpanSuccess, endSpanError, } from "./spans.js";
 import { recordMcpToolCall, recordMcpToolError, recordMcpResourceRead, recordMcpSessionStart, } from "./metrics.js";
 const DEFAULT_MAX_PAYLOAD_SIZE = 4096;
+/** Detect MCP transport type from server internals. */
+function detectTransport(server) {
+    // Check if server has a transport reference
+    const transport = server.transport ?? server._transport;
+    if (transport && typeof transport === "object") {
+        const name = transport.constructor?.name ?? "";
+        if (/stdio/i.test(name))
+            return "pipe";
+        if (/sse|http/i.test(name))
+            return "tcp";
+    }
+    // Default: stdio is most common for local dev
+    return "pipe";
+}
 function truncate(value, maxBytes) {
     if (value.length <= maxBytes)
         return value;
@@ -75,7 +90,19 @@ server, options = {}) {
     const redactKeys = options.redactKeys ?? [];
     const maxPayload = options.maxPayloadSize ?? DEFAULT_MAX_PAYLOAD_SIZE;
     const propagate = options.propagateContext ?? true;
-    const spanOpts = { serverName, serverVersion };
+    // Session ID: explicit > server-provided > generated
+    const sessionId = options.sessionId ?? server.sessionId ?? randomUUID().slice(0, 8);
+    // Protocol version from server capabilities or default
+    const protocolVersion = server.protocolVersion ?? server._protocolVersion ?? "2025-03-26";
+    // Detect transport type: pipe (stdio) vs tcp (HTTP)
+    const networkTransport = detectTransport(server);
+    const spanOpts = {
+        serverName,
+        serverVersion,
+        sessionId,
+        protocolVersion,
+        networkTransport,
+    };
     // --- Wrap .tool() ---
     const originalTool = server.tool.bind(server);
     server.tool = function wrappedTool(name, ...rest) {
@@ -227,6 +254,9 @@ export async function traceSampling(fn, options) {
         serverName: options.serverName ?? "mcp-server",
         serverVersion: options.serverVersion ?? "unknown",
         parentContext: options.parentContext,
+        sessionId: options.sessionId,
+        protocolVersion: options.protocolVersion,
+        networkTransport: options.networkTransport,
     });
     if (options.maxTokens !== undefined) {
         span.setAttribute("gen_ai.request.max_tokens", options.maxTokens);

package/dist/mcp/spans.d.ts

@@ -1,18 +1,28 @@
 /**
  * OTel span creation for MCP operations.
  *
- * Maps MCP methods to OTel GenAI semantic conventions:
- * - tools/call              → execute_tool {tool_name}
- * - resources/read          → retrieval {uri}
- * - prompts/get             → prompt {name}
- * - sampling/createMessage  → chat {model}
+ * Follows OTel MCP semantic conventions (PR #2083):
+ * - tools/call              → `tools/call {tool_name}`
+ * - resources/read          → `resources/read {uri}`
+ * - prompts/get             → `prompts/get {name}`
+ * - sampling/createMessage  → `sampling/createMessage {model}`
  */
 import { type Context } from "@opentelemetry/api";
 declare const tracer: import("@opentelemetry/api").Tracer;
+/** MCP method names per OTel semconv. */
+export declare const MCP_METHODS: {
+    readonly TOOLS_CALL: "tools/call";
+    readonly RESOURCES_READ: "resources/read";
+    readonly PROMPTS_GET: "prompts/get";
+    readonly SAMPLING_CREATE_MESSAGE: "sampling/createMessage";
+};
 export interface SpanOptions {
     readonly serverName: string;
     readonly serverVersion: string;
     readonly parentContext: Context;
+    readonly sessionId?: string | undefined;
+    readonly protocolVersion?: string | undefined;
+    readonly networkTransport?: string | undefined;
 }
 export declare function startToolSpan(toolName: string, options: SpanOptions): import("@opentelemetry/api").Span;
 export declare function startResourceSpan(uri: string, options: SpanOptions): import("@opentelemetry/api").Span;
@@ -21,6 +31,9 @@ export interface SamplingSpanOptions {
     readonly serverName: string;
     readonly serverVersion: string;
     readonly parentContext?: Context | undefined;
+    readonly sessionId?: string | undefined;
+    readonly protocolVersion?: string | undefined;
+    readonly networkTransport?: string | undefined;
 }
 export declare function startSamplingSpan(model: string, options: SamplingSpanOptions): import("@opentelemetry/api").Span;
 export declare function endSpanSuccess(span: ReturnType<typeof tracer.startSpan>): void;

package/dist/mcp/spans.js

@@ -1,56 +1,71 @@
 /**
  * OTel span creation for MCP operations.
  *
- * Maps MCP methods to OTel GenAI semantic conventions:
- * - tools/call              → execute_tool {tool_name}
- * - resources/read          → retrieval {uri}
- * - prompts/get             → prompt {name}
- * - sampling/createMessage  → chat {model}
+ * Follows OTel MCP semantic conventions (PR #2083):
+ * - tools/call              → `tools/call {tool_name}`
+ * - resources/read          → `resources/read {uri}`
+ * - prompts/get             → `prompts/get {name}`
+ * - sampling/createMessage  → `sampling/createMessage {model}`
  */
 import { trace, SpanStatusCode, SpanKind, } from "@opentelemetry/api";
 const TRACER_NAME = "toad-eye-mcp";
 const tracer = trace.getTracer(TRACER_NAME);
+/** MCP method names per OTel semconv. */
+export const MCP_METHODS = {
+    TOOLS_CALL: "tools/call",
+    RESOURCES_READ: "resources/read",
+    PROMPTS_GET: "prompts/get",
+    SAMPLING_CREATE_MESSAGE: "sampling/createMessage",
+};
+/** Build common MCP attributes shared across all span types. */
+function baseAttrs(method, options) {
+    const attrs = {
+        "gen_ai.operation.name": method,
+        "mcp.method.name": method,
+        "mcp.server.name": options.serverName,
+        "mcp.server.version": options.serverVersion,
+    };
+    if (options.sessionId)
+        attrs["mcp.session.id"] = options.sessionId;
+    if (options.protocolVersion)
+        attrs["mcp.protocol.version"] = options.protocolVersion;
+    if (options.networkTransport)
+        attrs["network.transport"] = options.networkTransport;
+    return attrs;
+}
 export function startToolSpan(toolName, options) {
-    return tracer.startSpan(`execute_tool ${toolName}`, {
+    return tracer.startSpan(`${MCP_METHODS.TOOLS_CALL} ${toolName}`, {
         kind: SpanKind.INTERNAL,
         attributes: {
-            "gen_ai.operation.name": "execute_tool",
+            ...baseAttrs(MCP_METHODS.TOOLS_CALL, options),
             "gen_ai.tool.name": toolName,
-            "mcp.server.name": options.serverName,
-            "mcp.server.version": options.serverVersion,
         },
     }, options.parentContext);
 }
 export function startResourceSpan(uri, options) {
-    return tracer.startSpan(`retrieval ${uri}`, {
+    return tracer.startSpan(`${MCP_METHODS.RESOURCES_READ} ${uri}`, {
         kind: SpanKind.INTERNAL,
         attributes: {
-            "gen_ai.operation.name": "retrieval",
+            ...baseAttrs(MCP_METHODS.RESOURCES_READ, options),
             "gen_ai.data_source.id": uri,
-            "mcp.server.name": options.serverName,
-            "mcp.server.version": options.serverVersion,
         },
     }, options.parentContext);
 }
 export function startPromptSpan(promptName, options) {
-    return tracer.startSpan(`prompt ${promptName}`, {
+    return tracer.startSpan(`${MCP_METHODS.PROMPTS_GET} ${promptName}`, {
         kind: SpanKind.INTERNAL,
         attributes: {
-            "gen_ai.operation.name": "prompt",
+            ...baseAttrs(MCP_METHODS.PROMPTS_GET, options),
             "gen_ai.prompt.name": promptName,
-            "mcp.server.name": options.serverName,
-            "mcp.server.version": options.serverVersion,
         },
     }, options.parentContext);
 }
 export function startSamplingSpan(model, options) {
-    return tracer.startSpan(`chat ${model}`, {
+    return tracer.startSpan(`${MCP_METHODS.SAMPLING_CREATE_MESSAGE} ${model}`, {
         kind: SpanKind.CLIENT,
         attributes: {
-            "gen_ai.operation.name": "chat",
+            ...baseAttrs(MCP_METHODS.SAMPLING_CREATE_MESSAGE, options),
             "gen_ai.request.model": model,
-            "mcp.server.name": options.serverName,
-            "mcp.server.version": options.serverVersion,
         },
     }, options.parentContext);
 }
@@ -63,5 +78,10 @@ export function endSpanError(span, error) {
     const errorType = error instanceof Error ? error.constructor.name : "UnknownError";
     span.setStatus({ code: SpanStatusCode.ERROR, message });
     span.setAttribute("error.type", errorType);
+    // JSON-RPC error code if available (MCP SDK errors often carry a code)
+    const code = error?.code;
+    if (typeof code === "number") {
+        span.setAttribute("rpc.response.status_code", code);
+    }
     span.end();
 }

package/dist/mcp/types.d.ts

@@ -13,6 +13,8 @@ export interface ToadMcpOptions {
     readonly redactKeys?: readonly string[] | undefined;
     /** Max payload size in bytes before truncation. Default: 4096. */
     readonly maxPayloadSize?: number | undefined;
+    /** Explicit session ID. Auto-generated if not provided. */
+    readonly sessionId?: string | undefined;
     /** Extract W3C traceparent from _meta field. Default: true. */
     readonly propagateContext?: boolean | undefined;
     /** Hook called on each tool call span. Add custom attributes here. */
@@ -23,12 +25,15 @@ export interface ToadMcpOptions {
 /** OTel attributes set on MCP spans. */
 export interface McpSpanAttributes {
     readonly "gen_ai.operation.name": string;
+    readonly "mcp.method.name": string;
     readonly "gen_ai.tool.name"?: string;
     readonly "gen_ai.tool.call.id"?: string;
     readonly "gen_ai.data_source.id"?: string;
     readonly "gen_ai.prompt.name"?: string;
     readonly "mcp.server.name": string;
     readonly "mcp.server.version": string;
-    readonly "mcp.transport"?: "stdio" | "sse" | "streamable-http";
     readonly "mcp.session.id"?: string;
+    readonly "mcp.protocol.version"?: string;
+    readonly "network.transport"?: "pipe" | "tcp";
+    readonly "rpc.response.status_code"?: number;
 }

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "toad-eye",
-  "version": "2.6.0",
-  "description": "Auto-instrumentation, cost tracking, budget guards, and observability for LLM services",
+  "version": "2.7.0",
+  "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",
   "types": "dist/index.d.ts",
@@ -49,7 +49,7 @@
     "build": "tsc -p tsconfig.build.json",
     "test": "vitest run",
     "test:watch": "vitest",
-    "prepublishOnly": "npm run build"
+    "prepublishOnly": "test -s README.md || { echo '❌ README.md is missing or empty'; exit 1; } && npm run build"
   },
   "repository": {
     "type": "git",
@@ -58,6 +58,11 @@
   "homepage": "https://github.com/vola-trebla/toad-eye#readme",
   "author": "albertalov",
   "keywords": [
+    "mcp",
+    "model-context-protocol",
+    "mcp-server",
+    "mcp-observability",
+    "mcp-monitoring",
     "opentelemetry",
     "llm",
     "instrumentation",
@@ -71,9 +76,13 @@
     "cost-tracking",
     "auto-instrumentation",
     "gen-ai",
+    "ai-agent",
+    "tool-calling",
     "budget-guards",
     "finops",
-    "streaming"
+    "streaming",
+    "jaeger",
+    "prometheus"
   ],
   "license": "ISC",
   "dependencies": {

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

@@ -23,7 +23,7 @@
       },
       "targets": [
         {
-          "expr": "sum(rate(gen_ai_mcp_tool_calls_total[5m]))",
+          "expr": "sum(rate(gen_ai_mcp_tool_calls_total{gen_ai_tool_name=~\"$tool_name\"}[5m]))",
           "refId": "A"
         }
       ],
@@ -66,7 +66,7 @@
       },
       "targets": [
         {
-          "expr": "sum(rate(gen_ai_mcp_tool_duration_sum[5m])) / sum(rate(gen_ai_mcp_tool_duration_count[5m])) or vector(0)",
+          "expr": "sum(rate(gen_ai_mcp_tool_duration_sum{gen_ai_tool_name=~\"$tool_name\"}[5m])) / sum(rate(gen_ai_mcp_tool_duration_count{gen_ai_tool_name=~\"$tool_name\"}[5m])) or vector(0)",
           "refId": "A"
         }
       ],
@@ -109,7 +109,7 @@
       },
       "targets": [
         {
-          "expr": "sum(rate(gen_ai_mcp_tool_errors_total[5m])) / sum(rate(gen_ai_mcp_tool_calls_total[5m])) or vector(0)",
+          "expr": "sum(rate(gen_ai_mcp_tool_errors_total{gen_ai_tool_name=~\"$tool_name\"}[5m])) / sum(rate(gen_ai_mcp_tool_calls_total{gen_ai_tool_name=~\"$tool_name\"}[5m])) or vector(0)",
           "refId": "A"
         }
       ],
@@ -187,7 +187,7 @@
       },
       "targets": [
         {
-          "expr": "sum by (gen_ai_tool_name) (rate(gen_ai_mcp_tool_calls_total[5m]))",
+          "expr": "sum by (gen_ai_tool_name) (rate(gen_ai_mcp_tool_calls_total{gen_ai_tool_name=~\"$tool_name\"}[5m]))",
           "legendFormat": "{{gen_ai_tool_name}}",
           "refId": "A"
         }
@@ -231,12 +231,12 @@
       },
       "targets": [
         {
-          "expr": "histogram_quantile(0.50, sum by (gen_ai_tool_name, le) (rate(gen_ai_mcp_tool_duration_bucket[5m])))",
+          "expr": "histogram_quantile(0.50, sum by (gen_ai_tool_name, le) (rate(gen_ai_mcp_tool_duration_bucket{gen_ai_tool_name=~\"$tool_name\"}[5m])))",
           "legendFormat": "p50 {{gen_ai_tool_name}}",
           "refId": "A"
         },
         {
-          "expr": "histogram_quantile(0.95, sum by (gen_ai_tool_name, le) (rate(gen_ai_mcp_tool_duration_bucket[5m])))",
+          "expr": "histogram_quantile(0.95, sum by (gen_ai_tool_name, le) (rate(gen_ai_mcp_tool_duration_bucket{gen_ai_tool_name=~\"$tool_name\"}[5m])))",
           "legendFormat": "p95 {{gen_ai_tool_name}}",
           "refId": "B"
         }
@@ -280,7 +280,7 @@
       },
       "targets": [
         {
-          "expr": "sum by (gen_ai_tool_name, error_type) (rate(gen_ai_mcp_tool_errors_total[5m]))",
+          "expr": "sum by (gen_ai_tool_name, error_type) (rate(gen_ai_mcp_tool_errors_total{gen_ai_tool_name=~\"$tool_name\"}[5m]))",
           "legendFormat": "{{gen_ai_tool_name}} ({{error_type}})",
           "refId": "A"
         }
@@ -371,28 +371,28 @@
       },
       "targets": [
         {
-          "expr": "sum by (gen_ai_tool_name) (rate(gen_ai_mcp_tool_calls_total[5m]))",
+          "expr": "sum by (gen_ai_tool_name) (rate(gen_ai_mcp_tool_calls_total{gen_ai_tool_name=~\"$tool_name\"}[5m]))",
           "legendFormat": "",
           "refId": "calls",
           "format": "table",
           "instant": true
         },
         {
-          "expr": "sum by (gen_ai_tool_name) (rate(gen_ai_mcp_tool_duration_sum[5m])) / sum by (gen_ai_tool_name) (rate(gen_ai_mcp_tool_duration_count[5m]))",
+          "expr": "sum by (gen_ai_tool_name) (rate(gen_ai_mcp_tool_duration_sum{gen_ai_tool_name=~\"$tool_name\"}[5m])) / sum by (gen_ai_tool_name) (rate(gen_ai_mcp_tool_duration_count{gen_ai_tool_name=~\"$tool_name\"}[5m]))",
           "legendFormat": "",
           "refId": "avg_duration",
           "format": "table",
           "instant": true
         },
         {
-          "expr": "histogram_quantile(0.95, sum by (gen_ai_tool_name, le) (rate(gen_ai_mcp_tool_duration_bucket[5m])))",
+          "expr": "histogram_quantile(0.95, sum by (gen_ai_tool_name, le) (rate(gen_ai_mcp_tool_duration_bucket{gen_ai_tool_name=~\"$tool_name\"}[5m])))",
           "legendFormat": "",
           "refId": "p95_duration",
           "format": "table",
           "instant": true
         },
         {
-          "expr": "sum by (gen_ai_tool_name) (rate(gen_ai_mcp_tool_errors_total[5m])) / sum by (gen_ai_tool_name) (rate(gen_ai_mcp_tool_calls_total[5m]))",
+          "expr": "sum by (gen_ai_tool_name) (rate(gen_ai_mcp_tool_errors_total{gen_ai_tool_name=~\"$tool_name\"}[5m])) / sum by (gen_ai_tool_name) (rate(gen_ai_mcp_tool_calls_total{gen_ai_tool_name=~\"$tool_name\"}[5m]))",
           "legendFormat": "",
           "refId": "error_rate",
           "format": "table",
@@ -505,7 +505,46 @@
   "schemaVersion": 39,
   "tags": ["toad-eye", "mcp", "llm"],
   "templating": {
-    "list": []
+    "list": [
+      {
+        "name": "mcp_method",
+        "label": "MCP Method",
+        "type": "query",
+        "datasource": {
+          "type": "prometheus",
+          "uid": "prometheus"
+        },
+        "query": "label_values(gen_ai_mcp_tool_calls_total, mcp_method_name)",
+        "refresh": 2,
+        "includeAll": true,
+        "allValue": ".*",
+        "current": {
+          "text": "All",
+          "value": "$__all"
+        },
+        "multi": true,
+        "sort": 1
+      },
+      {
+        "name": "tool_name",
+        "label": "Tool",
+        "type": "query",
+        "datasource": {
+          "type": "prometheus",
+          "uid": "prometheus"
+        },
+        "query": "label_values(gen_ai_mcp_tool_calls_total, gen_ai_tool_name)",
+        "refresh": 2,
+        "includeAll": true,
+        "allValue": ".*",
+        "current": {
+          "text": "All",
+          "value": "$__all"
+        },
+        "multi": true,
+        "sort": 1
+      }
+    ]
   },
   "time": {
     "from": "now-1h",

package/templates/otel-collector.yml

@@ -31,7 +31,7 @@ processors:
           sampling_percentage: 100
 
 exporters:
-  otlp_http/jaeger:
+  otlphttp/jaeger:
     endpoint: http://jaeger:4318
     tls:
       insecure: true
@@ -44,7 +44,7 @@ service:
     traces:
       receivers: [otlp]
       processors: [batch]
-      exporters: [otlp_http/jaeger]
+      exporters: [otlphttp/jaeger]
     metrics:
       receivers: [otlp]
       processors: [batch]