API Telemetry

Overview

The framework auto-instruments every tool, resource, prompt, storage, LLM, speech, and graph call — each gets its own span and the standard counters/histograms. HTTP server requests pick up spans from HttpInstrumentation (all Node.js HTTP traffic, skips /healthz) plus httpInstrumentationMiddleware from @hono/otel on the MCP HTTP endpoint when installed (optional Tier 3 peer — bun add @hono/otel). On Bun, HttpInstrumentation silently no-ops and @hono/otel is the only HTTP coverage. Auth checks and session lifecycle are tracked as metrics only — auth decorates the active HTTP span with attributes, sessions emit counters.

requestId, traceId, and tenantId correlate automatically across spans, metrics, and logs. Pino logs get trace_id/span_id injected when a span is active.

A handler's ctx.traceId / ctx.spanId name the execution span it runs in — tool_execution:<name> or resource_read:<name> — not the enclosing HTTP request span. Under HTTP the trace ID is the request's, so handler logs join to the request; the span ID is the child execution's, so they join to that span's attributes and duration. On stdio, where no transport span exists, both are still populated from the execution span the framework opens. Both are undefined when telemetry is disabled: the non-recording span a disabled pipeline produces carries all-zero IDs, and the framework reports no correlation rather than IDs that correlate to nothing.

For the helper API surface (withSpan, createCounter, createHistogram, buildTraceparent, etc.) — see the api-utils skill, Telemetry section. This skill is the catalog of what is emitted; that one is the reference for how to emit your own.

Enabling export

OTel is off by default. OTEL_ENABLED=true alone does nothing — you also need an OTLP endpoint. Without an endpoint the SDK is configured but nothing leaves the process.

Metrics push via PeriodicExportingMetricReader every 15 seconds. Traces use BatchSpanProcessor.

Runtime support

Cloud platform detection auto-populates resource attributes:

Flush at exit

Spans batch and metrics push on a 15-second cycle, so a process that exits between cycles takes its telemetry with it. ServerHandle.shutdown() is the drain: it stops the transport, then force-flushes traces and metrics through the OTLP exporters and closes the logger.

Stdin EOF is a disconnect. A stdio host closing the pipe runs the cleanup a signal runs, exactly once — the shutdown detaches the signal handlers and the EOF watcher as it starts, so neither can re-enter it — and the process then exits explicitly instead of waiting to run out of handles. Two things follow: the OTLP export leaves the process, and a setInterval a service registered without unref() can no longer keep the server resident after its client is gone. The path writes nothing to stdout.

The drain is bounded. Shutdown-on-exit races a 10-second backstop, so a cleanup step that never settles still terminates the process. The logger bounds its own flush separately, per pino instance: a completing callback is awaited in full, and a runtime whose callback never arrives releases shutdown rather than hanging it.

Workers has no ServerHandle and no NodeSDK — flush whatever exporter you wired there yourself, via ctx.waitUntil().

Spans

Every handler call gets a span. Nested operations (storage, graph, LLM) become child spans on the same trace. All spans carry code.function.name and code.namespace for code-attribution. Errors are recorded via span.recordException() and SpanStatusCode.ERROR; McpError codes surface as the *.error_code attribute.

A handler that ends its round with ctx.requestInput(...) closes its span OK with mcp.*.input_required set — no recorded exception, no error-counter increment. Multi-round-trip input is protocol control flow, so it never inflates error rates; split on that attribute to tell an incomplete round from a completed call.

The measured region

A tool or resource call is measured from the start of the handler through the response pipeline that follows it: output-schema validation, format(), the enrichment merge, and the trailer render for tools; output-schema validation and format() for resources. Telemetry therefore records the outcome the client sees — a failure in any of those is an ERROR span, success=false counters, an error-counter increment, and isSuccess: false in the completion log, matching the isError: true the caller receives. Prompt generation has no post-handler pipeline, so its region is the generate function alone.

Two consequences worth knowing when reading a dashboard:

mcp.tool.partial_success and the mcp.tool.batch.* counts read the same domain value, so a batch envelope ({ succeeded, failed }) is still detected once the result has been assembled around it.

Trace context propagates across boundaries via W3C traceparent headers. See api-utils → telemetry/trace for withSpan, buildTraceparent, extractTraceparent, createContextWithParentTrace, injectCurrentContextInto, runInContext signatures.

Metrics

All custom metrics are namespaced mcp.* (or process.* / http.client.* where standard semconv applies). Lazy-initialized on first emission; tool, resource, prompt, http.client.request.duration, heartbeat, session, auth, rate-limit, and error metrics are eagerly created at startup so series exist from the first export cycle. LLM, speech, graph, and storage instruments are lazy-initialized on first use.

Tools, resources, prompts

Storage, LLM, speech, graph

Transport, auth, sessions

Errors, rate limits, HTTP client

Process

Auto-registered when process.memoryUsage / process.uptime / perf_hooks are available (Node/Bun, not Workers). The three memory gauges share a single process.memoryUsage() snapshot per collection cycle, refreshed at most every 100 ms.

Logs

Pino logs are auto-instrumented by @opentelemetry/instrumentation-pino. When a span is active, trace_id and span_id are injected into the record. Combined with the framework logger's automatic requestId/tenantId correlation, every log line is searchable by trace.

For domain logging inside handlers, use ctx.log (debug/info/notice/warning/error) — auto-includes requestId, traceId, tenantId, spanId. The completion log emitted at the end of every handler carries a metrics payload, with fields tuned to each surface:

Custom instrumentation

Need a span or metric for your own service? Use the helpers from @cyanheads/mcp-ts-core/utils (full signatures in api-utils → Telemetry):

import { withSpan, createCounter, createHistogram } from '@cyanheads/mcp-ts-core/utils';

const myOps = createCounter('myservice.operations', 'My service ops', '{ops}');
const myDuration = createHistogram('myservice.duration', 'My service duration', 'ms');

export async function doWork() {
  return withSpan('myservice.do_work', async (span) => {
    const t0 = performance.now();
    try {
      const result = await reallyDoWork();
      span.setAttribute('myservice.items', result.length);
      return result;
    } finally {
      myDuration.record(performance.now() - t0);
      myOps.add(1);
    }
  }, { 'myservice.region': 'us-west' });
}

Span context propagates automatically — withSpan calls inside a tool_execution:* span appear as children. runInContext(ctx, fn) carries the active OTel context across async boundaries (setTimeout, queueMicrotask).

For attribute keys, prefer the ATTR_* constants exported from @cyanheads/mcp-ts-core/utils (telemetry/attributes) over hand-typed strings — keeps you in step with framework conventions and avoids typos. Standard OTel semantic conventions (HTTP, cloud, service, network, etc.) are NOT re-exported — import those directly from @opentelemetry/semantic-conventions.

Visualization

An example Grafana dashboard JSON and vendor-agnostic query recipes (Prometheus, Datadog, New Relic, Honeycomb) live at docs/telemetry/ in the framework source — not bundled in the npm package, so consult the GitHub repo.

Cardinality discipline

Series are cheap to emit but expensive to store and query. The framework deliberately keeps high-cardinality identifiers off metric attributes and on spans only. Follow the same rule when adding your own metrics.

When in doubt: if the attribute can take more than ~100 distinct values across a fleet's runtime, it belongs on the span, not the metric.