| name | otel |
| description | Query and debug OpenTelemetry data (traces, logs, metrics) in ClickHouse (HyperDX/ClickStack) via the sideclaw MCP `otel` tool. Use when investigating application errors, slow or missing traces, log anomalies, service health, or any observability question in local dev or VPS production (local + prod both supported).
|
OTEL Debug — via sideclaw MCP
Call mcp__sideclaw__otel with:
investigation — what to look into (error message, service, trace id, anomaly, time range).environment — local or prod.cwd (optional) — defaults to $HOME. OTEL access is host-level, so this rarely matters.
The worker runs read-only on claude-sonnet-5[1m], currently on Max per
SIDECLAW_WORKER_BACKEND=max (IU unified endpoint fallback when unset — non-EU
routing, so treat prod log content accordingly), and queries ClickHouse through
~/.claude/skills/otel/scripts/query.py. Only the structured result (status,
environment, timeRange, findings, recommendations) returns to the caller —
the raw query output stays in the worker.
Inspect status first: errors = active error spans/logs, degraded = elevated
latency or warnings, healthy = data flowing normally.
Reliability: a query can take 1–6 min under load. The tool's 8-min timeout
absorbs most of it; on a hard timeout, retry.
Maintenance: the ClickHouse query script (scripts/query.py) is the canonical
access path and lives here in dotfiles, not in sideclaw — add presets / schema
changes there. The sideclaw worker invokes it by absolute $HOME path.
