[Bug] OTEL Export functionality disabled or non-functional

Additional data point — the regression also affects the http/json OTLP transport, and on a later version than reported here.

Setup: macOS, local OTLP collector on http://localhost:4318, config via ~/.claude/settings.json env block (also reproduced with the vars exported directly in the shell):

CLAUDE_CODE_ENABLE_TELEMETRY=1
OTEL_METRICS_EXPORTER=otlp
OTEL_LOGS_EXPORTER=otlp
OTEL_EXPORTER_OTLP_PROTOCOL=http/json
OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318

A/B test, same machine, same healthy collector (manual POST /v1/logs → HTTP 200):

| Version | OTLP (http/json) export |
|---------|-------------------------|
| 2.1.185 | ✅ logs + metrics received |
| 2.1.191 → 2.1.195 | ❌ zero requests, zero errors, nothing on stderr |

So unlike the http/protobuf exporter (reported broken since 2.1.113), the http/json exporter kept working through 2.1.185 and broke somewhere in 2.1.186–2.1.191. This narrows the bundling regression: both OTLP transports are now no-ops, but they dropped at different versions. OTEL_METRICS_EXPORTER=console is unaffected, consistent with the OTLP exporter packages being excluded from the compiled binary.

Closing for now — inactive for too long. Please open a new issue if this is still relevant.

I tried to reproduce this on Linux with the released 2.1.233, and also on 2.1.179 (your version) and 2.1.111 (your known-good pin), following https://code.claude.com/docs/en/monitoring-usage with a local OTLP receiver on localhost:4318.

Result: OTLP export worked on every version — http/json and http/protobuf both delivered real POSTs to /v1/metrics and /v1/logs, and grpc opened real gRPC connections to :4317. So the OTLP exporters are still compiled into the binary, including the native builds introduced in 2.1.113.

I did find one way to get exactly your symptom (zero output, zero errors): if OTEL_EXPORTER_OTLP_PROTOCOL (or the per-signal protocol variable) is unset, telemetry initialization fails silently — Claude Code has no default OTLP protocol (docs), and the error is only visible with --debug. This behaves the same on 2.1.111, though, so it wouldn't explain a change at 2.1.113 by itself. Also note that on machines with managed settings, OTLP variables set there remove developer-set per-signal variables at startup (docs).

To dig further, could you share:

1. The exact OTEL_* variables you set (and where — shell vs managed-settings.json env block)
2. The telemetry-related lines from claude --debug (they're prefixed [3P telemetry])
3. Whether adding OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf changes anything

🤖 Generated with Claude Code

We weren't able to reproduce this. Could you provide steps to trigger the issue — what you ran, what happened, and what you expected? This issue will be closed automatically if there's no activity within 7 days.