Skip to main content

Documentation Index

Fetch the complete documentation index at: https://sedataai.mintlify.app/llms.txt

Use this file to discover all available pages before exploring further.

​
I don’t see any traces in the dashboard

Walk this list top to bottom — the first match is almost always the cause.
Tools registered before instrumentation are not patched. Move instrumentServer(...) to immediately after new McpServer(...).
samplingRate: 0 drops everything. Default is 1.0. Print it on startup:
console.log('samplingRate =', telemetryConfig.samplingRate ?? 1.0)
enableTracing: false disables the exporter entirely. Default is true.
Should look like https://otel.sedata-ai.tech/v1. The package appends /traces and /metrics. A trailing slash is fine. A path like /v1/traces would result in /v1/traces/traces — wrong.
Switch to console exporter and confirm spans are produced at all:
exporterType: 'console'
If they print, the issue is auth or endpoint. If they don’t print, the issue is upstream of export (instrumentation or sampler).
Run npm ls @opentelemetry/api. Multiple versions create no-op tracers. Dedupe with npm dedupe or pin a single version.
Short-lived scripts can exit before the batch processor flushes. Always await telemetry.shutdown() before process.exit.

​
I see traces but metrics are missing

The console exporter has no metric reader — by design, for local debugging. Switch to otlp-http to get metrics.
Default is true. If your config object came from another module, double check it isn’t being overridden.
Default is 5000 ms. If you set 60000, you’ll wait a minute between flushes. For local dev, try 2000.
Without it, the last batch — including mcp.server.session.duration — is dropped.

​
Safety checks are always failing

console.warn: Safety check API call failed, allowing execution
The wrapper uses the key passed to instrumentServer via exporterAuth.token or exporterAuth.apiKey. Confirm it’s loaded:
console.log('SEDATA_TOKEN set:', !!process.env.SEDATA_TOKEN)
The wrapper only checks string values. Numbers, objects, and undefined pass through untouched and no safety attributes are recorded.
parameterName must match a field on your inputSchema. Typos result in silent skips.
Try curl https://api.sedata-ai.tech/security/safety-check — if it can’t reach, neither can the SDK.

​
My handler runs even when content is flagged

The wrapper short-circuits on flagged content — it does not call your handler. If your handler runs anyway, one of these is true:
  • The wrapper isn’t actually applied (you registered the raw handler).
  • The safety API failed (logged warning) and the wrapper fail-opened.
  • The parameter wasn’t a string and was skipped.
Confirm by inspecting the span: a mcp.safety_check.flagged === true attribute means the wrapper ran. If you see it AND your handler ran, file an issue with the trace id.

​
The error spam in dev is annoying

The wrapper prints debug logs by default (e.g. Safety check wrapper called with params: ...). These are deliberate while the package is at 0.x. To silence: 
const origLog = console.log
console.log = (...args) => {
  if (typeof args[0] === 'string' && args[0].startsWith('Safety check')) return
  if (typeof args[0] === 'string' && args[0].startsWith('Instrumentation params')) return
  if (typeof args[0] === 'string' && args[0].startsWith('Has _safetyCheck')) return
  origLog.apply(console, args)
}
Track #logging for a proper log-level option.

​
TypeScript: type ‘any’ on tool params

The safetyCheck wrapper currently types its handler as (params: any) => any. To get proper typing, annotate the inner function: 
safetyCheck(
  async ({ text }: { text: string }) => { /* ... */ },
  { parameterName: 'text' },
)

​
Still stuck?

FAQ

Quick answers to common questions.

Support

How to reach the team.