Skip to main content
A Next.js dev server already emits OpenTelemetry spans, but the evidence is scattered across terminals, framework internals and database logs. StackNext collects the supported spans, query events and request errors into one bounded in-memory sample and gives you four ways to read it:
  • a dashboard served by the local collector;
  • stacknext CLI commands for status, coverage, diagnostics, snapshots and regression records;
  • a read-only MCP server for AI coding agents;
  • redacted snapshots and regression records stored under .stacknext.

​
Data flow

Fig 0.1Local data flow

Your app

  • Next.js
  • Drizzle
  • Prisma

plus custom spans and manual action helpers

spans · query events · errors

Instrumentation

redact

@stacknext/next

redacts in the app · sends asynchronously, off the request path

loopback only · bearer token

Collector · one Node process

redact

127.0.0.1:4400

validates every envelope · never blocks a request

in-memory sample

5,000 spans · 10-minute window

capacity limits · drop counters

same sample · read-only

Dashboard

requests · traces · queries

CLI + export

coverage · diagnose · snapshot

MCP server

read-only tools · GET only

only on an explicit save

Local records · redacted

snapshotsrunsbaselinescomparisons

.stacknext/ · 0600 files · never uploaded or auto-deleted

stacknext dev starts the collector and runs your app’s own dev script. It writes the collector URL and a short-lived bearer token to .stacknext/run.json and passes them to the Next.js process. The app sends telemetry over loopback only.

​
What you can answer

​
Local-first by design

Runtime evidence often contains enough information to identify a bug, but sending it to a hosted service adds an account, a network boundary and a new privacy decision. StackNext binds the collector to 127.0.0.1, requires a bearer token for ingestion, stores telemetry in memory with capacity and retention limits, and keeps explicit records in redacted local files. The result is a debugging loop that works offline and never needs an AI API key. The MCP client you connect may still send tool results to a hosted model; that is the client’s policy, not a StackNext network path.

​
What it deliberately does not do

  • It does not capture bodies, headers, cookies, URL query values or SQL bind parameters.
  • It does not support the Edge runtime, production instrumentation or frameworks other than Next.js App Router.
  • It does not run SQL or EXPLAIN, edit source code or recommend indexes from telemetry alone.
  • It does not treat a heuristic finding as a confirmed root cause or a comparison as a statistical test.
  • It does not replace an existing OpenTelemetry provider.
Read core concepts, limitations and privacy before relying on it.