sentry-node-sdk — скилл для Claude Code: Sentry в Node.js, Bun и Deno
Скилл сканирует проект и ведёт по шагам подключения Sentry к серверному JavaScript и TypeScript: ошибки, трейсинг, логи, кроны.
npx skills add getsentry/sentry-for-ai --skill sentry-node-sdk -a claude-codeЧто делает sentry-node-sdk
Скилл работает как мастер: сначала сам смотрит, что у вас за рантайм и стек, потом предлагает конкретный набор функций. Проверяет версии bun, deno, node, ищет в package.json пакеты @sentry/*, фреймворки вроде Express, Fastify, Koa, Hapi, Elysia, Hono, библиотеки логирования (winston, pino, bunyan), планировщики (node-cron, Bull, Agenda) и клиенты LLM. Отдельно выясняет, ESM у вас или CommonJS, и есть ли уже файл instrument.js или instrument.mjs — если есть, скилл вливает настройки в него, а не перезаписывает.
Дальше идёт рекомендация, а не допрос: базово предлагаются мониторинг ошибок и трейсинг, остальное — по обстоятельствам. Профилирование доступно только на Node.js, на Bun и Deno его нет. Если в проекте уже есть OpenTelemetry-трейсинг, скилл уводит на путь OTLP через otlpIntegration() и советует не выставлять tracesSampleRate — нативная трассировка и профилирование в этом случае не подключаются. Для NestJS и Next.js он отправляет к другим скиллам: sentry-nestjs-sdk и sentry-nextjs-sdk.
Перед реализацией стоит сверить версии пакетов с документацией Sentry: в тексте скилла они привязаны к моменту написания. ESM-проектам нужен запуск через --import ./instrument.mjs, для Bun — --preload, а в Deno пакет подключается как npm:@sentry/deno. Это те места, где установка чаще всего ломается.
Какие задачи решает
- подключить Sentry к серверному приложению на Node.js, Bun или Deno
- разобраться, нужен ли OTLP-путь, если в проекте уже есть OpenTelemetry
- добавить мониторинг ошибок в Express, Fastify, Koa или Hapi
- настроить отслеживание пропущенных и упавших задач по расписанию
- включить сбор логов из winston или pino с привязкой к трейсам
Примеры запросов агенту
- добавь Sentry в моё Node.js-приложение
- настрой @sentry/bun для этого проекта
- подключи мониторинг ошибок и трейсинг к Fastify
Как установить sentry-node-sdk
1Через Skills CLI (Vercel)
npx skills add getsentry/sentry-for-ai --skill sentry-node-sdk -a claude-code2Вручную: положить файлы
~/.claude/skills/sentry-node-sdk/SKILL.mdПапка скилла целиком; для одного проекта — .claude/skills/sentry-node-sdk/
1Через Skills CLI (Vercel)
npx skills add getsentry/sentry-for-ai --skill sentry-node-sdk -a codex2Вручную: положить файлы
.agents/skills/sentry-node-sdk/SKILL.mdДля всех проектов — ~/.agents/skills/
1Через Skills CLI (Vercel)
npx skills add getsentry/sentry-for-ai --skill sentry-node-sdk -a cursor2Вручную: положить файлы
.cursor/skills/sentry-node-sdk/SKILL.mdCursor читает и .agents/skills/, и .claude/skills/
1GitHub CLI
gh skill install getsentry/sentry-for-ai sentry-node-sdk2Вручную: положить файлы
.github/skills/sentry-node-sdk/SKILL.md1Через Skills CLI (Vercel)
npx skills add getsentry/sentry-for-ai --skill sentry-node-sdk -a gemini-cli2Командой платформы
gemini skills install https://github.com/getsentry/sentry-for-ai --path skills-legacy/sentry-node-sdkВручную: положить файлы
.devin/skills/sentry-node-sdk/SKILL.mdСтарый путь .windsurf/skills/ тоже читается
Через Skills CLI (Vercel)
npx skills add getsentry/sentry-for-ai --skill sentry-node-sdk -a opencodeClaude.ai
Упакуйте папку в zip и загрузите в Settings → Capabilities → Skills
По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов
Вопросы о sentry-node-sdk
Как установить sentry-node-sdk в Claude?
Выполните npx skills add getsentry/sentry-for-ai --skill sentry-node-sdk -a claude-code. Скилл появится в ~/.claude/skills/sentry-node-sdk/SKILL.md.
Что делает sentry-node-sdk?
Сканирует проект, определяет рантайм и фреймворк, а затем ведёт по настройке Sentry: мониторинг ошибок, трейсинг, логи, профилирование, метрики, кроны и AI-мониторинг.
Чем этот скилл отличается от sentry-nestjs-sdk и sentry-nextjs-sdk?
Он рассчитан на серверные рантаймы Node.js, Bun и Deno. Для NestJS нужен sentry-nestjs-sdk с нативными декораторами, для Next.js — sentry-nextjs-sdk с тремя рантаймами.
Исходный текст
SKILL.md · Apache-2.0
---
name: sentry-node-sdk
description: Full Sentry SDK setup for Node.js, Bun, and Deno. Use when asked to "add Sentry to Node.js", "add Sentry to Bun", "add Sentry to Deno", "install @sentry/node", "@sentry/bun", or "@sentry/deno", or configure error monitoring, tracing, logging, profiling, metrics, crons, or AI monitoring for server-side JavaScript/TypeScript runtimes.
license: Apache-2.0
category: sdk-setup
parent: sentry-sdk-setup
disable-model-invocation: true
---
> [All Skills](../../SKILL_TREE.md) > [SDK Setup](../sentry-sdk-setup/SKILL.md) > Node.js / Bun / Deno SDK
# Sentry Node.js / Bun / Deno SDK
Opinionated wizard that scans your project and guides you through complete Sentry setup for server-side JavaScript and TypeScript runtimes: Node.js, Bun, and Deno.
## Invoke This Skill When
- User asks to "add Sentry to Node.js", "Bun", or "Deno"
- User wants to install or configure `@sentry/node`, `@sentry/bun`, or `@sentry/deno`
- User wants error monitoring, tracing, logging, profiling, crons, metrics, or AI monitoring for a backend JS/TS app
- User asks about `instrument.js`, `--import ./instrument.mjs`, `bun --preload`, or `npm:@sentry/deno`
- User wants to monitor Express, Fastify, Koa, Hapi, Connect, Bun.serve(), or Deno.serve()
> **NestJS?** Use [`sentry-nestjs-sdk`](../sentry-nestjs-sdk/SKILL.md) instead — it uses `@sentry/nestjs` with NestJS-native decorators and filters.
> **Next.js?** Use [`sentry-nextjs-sdk`](../sentry-nextjs-sdk/SKILL.md) instead — it handles the three-runtime architecture (browser, server, edge).
> **Note:** SDK versions below reflect current Sentry docs at time of writing (`@sentry/node` ≥10.42.0, `@sentry/bun` ≥10.42.0, `@sentry/deno` ≥10.42.0).
> Always verify against [docs.sentry.io/platforms/javascript/guides/node/](https://docs.sentry.io/platforms/javascript/guides/node/) before implementing.
---
## Phase 1: Detect
Run these commands to identify the runtime, framework, and existing Sentry setup:
```bash
# Detect runtime
bun --version 2>/dev/null && echo "Bun detected"
deno --version 2>/dev/null && echo "Deno detected"
node --version 2>/dev/null && echo "Node.js detected"
# Detect existing Sentry packages
cat package.json 2>/dev/null | grep -E '"@sentry/'
cat deno.json deno.jsonc 2>/dev/null | grep -i sentry
# Detect Node.js framework
cat package.json 2>/dev/null | grep -E '"express"|"fastify"|"@hapi/hapi"|"koa"|"@nestjs/core"|"connect"'
# Detect Bun-specific frameworks
cat package.json 2>/dev/null | grep -E '"elysia"|"hono"'
# Detect Deno frameworks (deno.json imports)
cat deno.json deno.jsonc 2>/dev/null | grep -E '"oak"|"hono"|"fresh"'
# Detect module system (Node.js)
cat package.json 2>/dev/null | grep '"type"'
ls *.mjs *.cjs 2>/dev/null | head -5
# Detect existing instrument file
ls instrument.js instrument.mjs instrument.ts instrument.cjs 2>/dev/null
# Detect logging libraries
cat package.json 2>/dev/null | grep -E '"winston"|"pino"|"bunyan"'
# Detect cron / scheduling
cat package.json 2>/dev/null | grep -E '"node-cron"|"cron"|"agenda"|"bull"|"bullmq"'
# Detect AI / LLM usage
cat package.json 2>/dev/null | grep -E '"openai"|"@anthropic-ai"|"@langchain"|"@vercel/ai"|"@google/generative-ai"'
# Detect OpenTelemetry tracing
cat package.json 2>/dev/null | grep -E '"@opentelemetry/sdk-node"|"@opentelemetry/sdk-trace-node"|"@opentelemetry/sdk-trace-base"'
grep -rn "NodeTracerProvider\|trace\.getTracer\|startActiveSpan" \
--include="*.ts" --include="*.js" --include="*.mjs" 2>/dev/null | head -5
# Check for companion frontend
ls frontend/ web/ client/ ui/ 2>/dev/null
cat package.json 2>/dev/null | grep -E '"react"|"vue"|"svelte"|"next"'
```
**What to determine:**
| Question | Impact |
|----------|--------|
| Which runtime? (Node.js / Bun / Deno) | Determines package, init pattern, and preload flag |
| Node.js: ESM or CJS? | ESM requires `--import ./instrument.mjs`; CJS uses `require("./instrument")` |
| Framework detected? | Determines which error handler to register |
| `@sentry/*` already installed? | Skip install, go straight to feature config |
| `instrument.js` / `instrument.mjs` already exists? | Merge into it rather than overwrite |
| Logging library detected? | Recommend Sentry Logs |
| Cron / job scheduler detected? | Recommend Crons monitoring |
| AI library detected? | Recommend AI Monitoring |
| OpenTelemetry tracing detected? | Use OTLP path instead of native tracing |
| Companion frontend found? | Trigger Phase 4 cross-link |
---
## Phase 2: Recommend
Present a concrete recommendation based on what you found. Don't ask open-ended questions — lead with a proposal:
**Route from OTel detection:**
- **OTel tracing detected** (`@opentelemetry/sdk-node` or `@opentelemetry/sdk-trace-node` in `package.json`, or `NodeTracerProvider` in source) → use OTLP path: `otlpIntegration()` via `@sentry/node-core/light`; do **not** set `tracesSampleRate`; Sentry links errors to OTel traces automatically
**Recommended (core coverage):**
- ✅ **Error Monitoring** — always; captures unhandled exceptions, promise rejections, and framework errors
- ✅ **Tracing** — automatic HTTP, DB, and queue instrumentation via OpenTelemetry
**Optional (enhanced observability):**
- ⚡ **Logging** — structured logs via `Sentry.logger.*`; recommend when `winston`/`pino`/`bunyan` or log search is needed
- ⚡ **Profiling** — continuous CPU profiling (Node.js only; not available on Bun or Deno); **not available with OTLP path**
- ⚡ **AI Monitoring** — OpenAI, Anthropic, LangChain, Vercel AI SDK; recommend when AI/LLM calls detected
- ⚡ **Crons** — detect missed or failed scheduled jobs; recommend when node-cron, Bull, or Agenda is detected
- ⚡ **Metrics** — custom counters, gauges, distributions; recommend when custom KPIs needed
- ⚡ **Runtime Metrics** — automatic collection of memory, CPU, and event loop metrics; `nodeRuntimeMetricsIntegration()` (Node.js) / `bunRuntimeMetricsIntegration()` (Bun)
**Recommendation logic:**
| Feature | Recommend when... |
|---------|------------------|
| Error Monitoring | **Always** — non-negotiable baseline |
| OTLP Integration | OTel tracing detected — **replaces** native Tracing |
| Tracing | **Always for server apps** — HTTP spans + DB spans are high-value; **skip if OTel tracing detected** |
| Logging | App uses winston, pino, bunyan, or needs log-to-trace correlation |
| Profiling | **Node.js only** — performance-critical service; native addon compatible; **skip if OTel tracing detected** (requires `tracesSampleRate`, incompatible with OTLP) |
| AI Monitoring | App calls OpenAI, Anthropic, LangChain, Vercel AI, or Google GenAI |
| Crons | App uses node-cron, Bull, BullMQ, Agenda, or any scheduled task pattern |
| Metrics | App needs custom counters, gauges, or histograms |
| Runtime Metrics | Any Node.js or Bun service wanting automatic memory/CPU/event-loop visibility |
**OTel tracing detected:** *"I see OpenTelemetry tracing in the project. I recommend Sentry's OTLP integration for tracing (via your existing OTel setup) + Error Monitoring + Sentry Logging [+ Metrics/Crons/AI Monitoring if applicable]. Shall I proceed?"*
**No OTel:** *"I recommend setting up Error Monitoring + Tracing. Want me to also add Logging or Profiling?"*
---
## Phase 3: Guide
### Runtime: Node.js
#### Option 1: Wizard (Recommended for Node.js)
> **You need to run this yourself** — the wizard opens a browser for login and requires interactive input that the agent can't handle. Copy-paste into your terminal:
>
> ```
> npx @sentry/wizard@latest -i node
> ```
>
> It handles login, org/project selection, SDK installation, `instrument.js` creation, and package.json script updates.
>
> **Once it finishes, come back and skip to [Verification](#verification).**
If the user skips the wizard, proceed with Option 2 (Manual Setup) below.
---
#### Option 2: Manual Setup — Node.js
##### Install
```bash
npm install @sentry/node --save
# or
yarn add @sentry/node
# or
pnpm add @sentry/node
```
##### Create the Instrument File
**CommonJS (`instrument.js`):**
```javascript
// instrument.js — must be loaded before all other modules
const Sentry = require("@sentry/node");
Sentry.init({
dsn: process.env.SENTRY_DSN ?? "___DSN___",
dataCollection: {
// To disable sending user data and HTTP bodies, uncomment the lines below. For more info visit:
// https://docs.sentry.io/platforms/javascript/guides/node/configuration/options/#dataCollection
// userInfo: false,
// httpBodies: [],
},
// 100% in dev, lower in production
tracesSampleRate: process.env.NODE_ENV === "development" ? 1.0 : 0.1,
// Capture local variable values in stack frames
includeLocalVariables: true,
enableLogs: true,
});
```
**ESM (`instrument.mjs`):**
```javascript
// instrument.mjs — loaded via --import flag before any other module
import * as Sentry from "@sentry/node";
Sentry.init({
dsn: process.env.SENTRY_DSN ?? "___DSN___",
dataCollection: {
// To disable sending user data and HTTP bodies, uncomment the lines below. For more info visit:
// https://docs.sentry.io/platforms/javascript/guides/node/configuration/options/#dataCollection
// userInfo: false,
// httpBodies: [],
},
tracesSampleRate: process.env.NODE_ENV === "development" ? 1.0 : 0.1,
includeLocalVariables: true,
enableLogs: true,
});
```
##### Start Your App with Sentry Loaded First
**CommonJS** — add `require("./instrument")` as the very first line of your entry file:
```javascript
// app.js
require("./instrument"); // must be first
const express = require("express");
// ... rest of your app
```
**ESM** — use the `--import` flag so Sentry loads before all other modules (Node.js 18.19.0+ required):
```bash
node --import ./instrument.mjs app.mjs
```
Add to `package.json` scripts:
```json
{
"scripts": {
"start": "node --import ./instrument.mjs server.mjs",
"dev": "node --import ./instrument.mjs --watch server.mjs"
}
}
```
Or via environment variable (useful for wrapping existing start commands):
```bash
NODE_OPTIONS="--import ./instrument.mjs" npm start
```
##### Framework Error Handlers
Register the Sentry error handler **after all routes** so it can capture framework errors:
**Express:**
```javascript
const express = require("express");
const Sentry = require("@sentry/node");
const app = express();
// ... your routes
// Add AFTER all routes — captures 5xx errors by default
Sentry.setupExpressErrorHandler(app);
// Optional: capture 4xx errors too
// Sentry.setupExpressErrorHandler(app, {
// shouldHandleError(error) { return error.status >= 400; },
// });
app.listen(3000);
```
**Fastify:**
```javascript
const Fastify = require("fastify");
const Sentry = require("@sentry/node");
const fastify = Fastify();
// Add BEFORE routes (unlike Express!)
Sentry.setupFastifyErrorHandler(fastify);
// ... your routes
await fastify.listen({ port: 3000 });
```
**Koa:**
```javascript
const Koa = require("koa");
const Sentry = require("@sentry/node");
const app = new Koa();
// Add as FIRST middleware (catches errors thrown by later middleware)
Sentry.setupKoaErrorHandler(app);
// ... your other middleware and routes
app.listen(3000);
```
**Hapi (async — must await):**
```javascript
const Hapi = require("@hapi/hapi");
const Sentry = require("@sentry/node");
const server = Hapi.server({ port: 3000 });
// ... your routes
// Must await — Hapi registration is async
await Sentry.setupHapiErrorHandler(server);
await server.start();
```
**Connect:**
```javascript
const connect = require("connect");
const Sentry = require("@sentry/node");
const app = connect();
// Add BEFORE routes (like Fastify and Koa)
Sentry.setupConnectErrorHandler(app);
// ... your middleware and routes
require("http").createServer(app).listen(3000);
```
**NestJS** — has its own dedicated skill with full coverage:
> **Use the [`sentry-nestjs-sdk`](../sentry-nestjs-sdk/SKILL.md) skill instead.**
> NestJS uses a separate package (`@sentry/nestjs`) with NestJS-native constructs:
> `SentryModule.forRoot()`, `SentryGlobalFilter`, `@SentryTraced`, `@SentryCron` decorators,
> and GraphQL/Microservices support. Load that skill for complete NestJS setup.
**Vanilla Node.js `http` module** — wrap request handler manually:
```javascript
const http = require("http");
const Sentry = require("@sentry/node");
const server = http.createServer((req, res) => {
Sentry.withIsolationScope(() => {
try {
// your handler
res.end("OK");
} catch (err) {
Sentry.captureException(err);
res.writeHead(500);
res.end("Internal Server Error");
}
});
});
server.listen(3000);
```
**Framework error handler summary:**
| Framework | Function | Placement | Async? |
|-----------|----------|-----------|--------|
| Express | `setupExpressErrorHandler(app)` | **After** all routes | No |
| Fastify | `setupFastifyErrorHandler(fastify)` | **Before** routes | No |
| Koa | `setupKoaErrorHandler(app)` | **First** middleware | No |
| Hapi | `setupHapiErrorHandler(server)` | Before `server.start()` | **Yes** |
| Connect | `setupConnectErrorHandler(app)` | **Before** routes | No |
| NestJS | → Use [`sentry-nestjs-sdk`](../sentry-nestjs-sdk/SKILL.md) | Dedicated skill | — |
---
### Runtime: Bun
> **No wizard available for Bun.** Manual setup only.
#### Install
```bash
bun add @sentry/bun
```
#### Create `instrument.ts` (or `instrument.js`)
```typescript
// instrument.ts
import * as Sentry from "@sentry/bun";
Sentry.init({
dsn: process.env.SENTRY_DSN ?? "___DSN___",
dataCollection: {
// To disable sending user data and HTTP bodies, uncomment the lines below. For more info visit:
// https://docs.sentry.io/platforms/javascript/guides/node/configuration/options/#dataCollection
// userInfo: false,
// httpBodies: [],
},
tracesSampleRate: process.env.NODE_ENV === "development" ? 1.0 : 0.1,
enableLogs: true,
});
```
#### Start Your App with `--preload`
```bash
bun --preload ./instrument.ts server.ts
```
Add to `package.json`:
```json
{
"scripts": {
"start": "bun --preload ./instrument.ts server.ts",
"dev": "bun --watch --preload ./instrument.ts server.ts"
}
}
```
#### Bun.serve() — Auto-Instrumentation
`@sentry/bun` automatically instruments `Bun.serve()` via JavaScript Proxy. No extra setup is required — just initialize with `--preload` and your `Bun.serve()` calls are traced:
```typescript
// server.ts
const server = Bun.serve({
port: 3000,
fetch(req) {
return new Response("Hello from Bun!");
},
});
```
#### Framework Error Handlers on Bun
Bun can run Express, Fastify, Hono, and Elysia. Use the same `@sentry/bun` import and the `@sentry/node` error handler functions (re-exported by `@sentry/bun`):
```typescript
import * as Sentry from "@sentry/bun";
import express from "express";
const app = express();
// ... routes
Sentry.setupExpressErrorHandler(app);
app.listen(3000);
```
#### Bun Feature Support
| Feature | Bun Support | Notes |
|---------|-------------|-------|
| Error Monitoring | ✅ Full | Same API as Node |
| Tracing | ✅ Via `@sentry/node` OTel | Most auto-instrumentations work |
| Logging | ✅ Full | `enableLogs: true` + `Sentry.logger.*` |
| Profiling | ❌ Not available | `@sentry/profiling-node` uses native addons incompatible with Bun |
| Metrics | ✅ Full | `Sentry.metrics.*` |
| Runtime Metrics | ✅ Full | `bunRuntimeMetricsIntegration()` — memory, CPU, event loop (no event loop delay percentiles) |
| Crons | ✅ Full | `Sentry.withMonitor()` |
| AI Monitoring | ✅ Full | OpenAI, Anthropic integrations work |
---
### Runtime: Deno
> **No wizard available for Deno.** Manual setup only.
> **Requires Deno 2.0+.** Deno 1.x is not supported.
> **Use `npm:` specifier.** The `deno.land/x/sentry` registry is deprecated.
#### Install via `deno.json` (Recommended)
```json
{
"imports": {
"@sentry/deno": "npm:@sentry/[email protected]"
}
}
```
Or import directly with the `npm:` specifier:
```typescript
import * as Sentry from "npm:@sentry/deno";
```
#### Initialize — Add to Entry File
```typescript
// main.ts — Sentry.init() must be called before any other code
import * as Sentry from "@sentry/deno";
Sentry.init({
dsn: Deno.env.get("SENTRY_DSN") ?? "___DSN___",
dataCollection: {
// To disable sending user data and HTTP bodies, uncomment the lines below. For more info visit:
// https://docs.sentry.io/platforms/javascript/guides/node/configuration/options/#dataCollection
// userInfo: false,
// httpBodies: [],
},
tracesSampleRate: Deno.env.get("DENO_ENV") === "development" ? 1.0 : 0.1,
enableLogs: true,
});
// Your application code follows
Deno.serve({ port: 8000 }, (req) => {
return new Response("Hello from Deno!");
});
```
> Unlike Node.js and Bun, Deno does not have a `--preload` or `--import` flag. Sentry must be the first `import` in your entry file.
#### Required Deno Permissions
The SDK requires network access to reach your Sentry ingest domain:
```bash
deno run \
--allow-net=o<ORG_ID>.ingest.sentry.io \
--allow-read=./src \
--allow-env=SENTRY_DSN,SENTRY_RELEASE \
main.ts
```
For development, `--allow-all` works but is not recommended for production.
#### Deno Cron Integration
Deno provides native cron scheduling. Use `denoCronIntegration` for automatic monitoring:
```typescript
import * as Sentry from "@sentry/deno";
import { denoCronIntegration } from "@sentry/deno";
Sentry.init({
dsn: Deno.env.get("SENTRY_DSN") ?? "___DSN___",
integrations: [denoCronIntegration()],
});
// Cron is automatically monitored
Deno.cron("daily-cleanup", "0 0 * * *", () => {
// cleanup logic
});
```
#### Deno Feature Support
| Feature | Deno Support | Notes |
|---------|-------------|-------|
| Error Monitoring | ✅ Full | Unhandled exceptions + `captureException` |
| Tracing | ✅ Custom OTel | Automatic spans for `Deno.serve()` and `fetch` |
| Logging | ✅ Full | `enableLogs: true` + `Sentry.logger.*` |
| Profiling | ❌ Not available | No profiling addon for Deno |
| Metrics | ✅ Full | `Sentry.metrics.*` |
| Runtime Metrics | ❌ Not available | No runtime metrics integration for Deno |
| Crons | ✅ Full | `denoCronIntegration()` + `Sentry.withMonitor()` |
| AI Monitoring | ✅ Partial | Vercel AI SDK integration works; OpenAI/Anthropic via `npm:` |
---
### OTLP Integration (OTel-First Projects — Node.js Only)
> Use this path **only when OpenTelemetry tracing was detected** in Phase 1
> (e.g., `@opentelemetry/sdk-node` or `@opentelemetry/sdk-trace-node` in `package.json`).
> For projects without an existing OTel setup, use the standard `@sentry/node` path above.
The OTLP integration uses `@sentry/node-core/light` — a lightweight Sentry SDK that does not bundle its own OpenTelemetry. Instead, it hooks into the user's existing OTel `TracerProvider` and exports spans to Sentry via OTLP.
#### When to Use
| Scenario | Recommended path |
|----------|-----------------|
| New project, no existing OTel | Standard `@sentry/node` (above) — includes built-in OTel |
| Existing OTel setup, want Sentry tracing | `@sentry/node-core/light` + `otlpIntegration()` |
| Existing OTel setup, sending to own Collector | `@sentry/node-core/light` + `otlpIntegration({ collectorUrl })` |
#### Install
```bash
npm install @sentry/node-core @opentelemetry/api @opentelemetry/sdk-trace-node @opentelemetry/sdk-trace-base
# or
yarn add @sentry/node-core @opentelemetry/api @opentelemetry/sdk-trace-node @opentelemetry/sdk-trace-base
# or
pnpm add @sentry/node-core @opentelemetry/api @opentelemetry/sdk-trace-node @opentelemetry/sdk-trace-base
```
> The `@opentelemetry/*` packages are peer dependencies. If the project already has them installed, skip duplicates.
#### Initialize
```javascript
// instrument.mjs — load via --import flag before any other module
import { NodeTracerProvider } from '@opentelemetry/sdk-trace-node';
import * as Sentry from '@sentry/node-core/light';
import { otlpIntegration } from '@sentry/node-core/light/otlp';
// Register the user's OTel TracerProvider first
const provider = new NodeTracerProvider();
provider.register();
Sentry.init({
dsn: process.env.SENTRY_DSN ?? '___DSN___',
dataCollection: {
// To disable sending user data and HTTP bodies, uncomment the lines below. For more info visit:
// https://docs.sentry.io/platforms/javascript/guides/node/configuration/options/#dataCollection
// userInfo: false,
// httpBodies: [],
},
enableLogs: true,
// Do NOT set tracesSampleRate — OTel controls sampling
integrations: [
otlpIntegration({
// Export OTel spans to Sentry via OTLP (default: true)
setupOtlpTracesExporter: true,
}),
],
});
```
**With a custom Collector endpoint:**
```javascript
Sentry.init({
dsn: process.env.SENTRY_DSN ?? '___DSN___',
integrations: [
otlpIntegration({
collectorUrl: 'http://localhost:4318/v1/traces',
}),
],
});
```
#### Start Your App
Same `--import` pattern as the standard Node.js setup:
```bash
node --import ./instrument.mjs app.mjs
```
#### Key Differences from Standard `@sentry/node`
| Aspect | `@sentry/node` (standard) | `@sentry/node-core/light` (OTLP) |
|--------|--------------------------|----------------------------------|
| OTel bundled | ✅ Yes — built-in TracerProvider | ❌ No — uses your existing provider |
| Tracing control | `tracesSampleRate` in `Sentry.init()` | OTel SDK controls sampling |
| Auto-instrumentation | ✅ Built-in (HTTP, DB, etc.) | ❌ You manage OTel instrumentations |
| Profiling | ✅ Available | ❌ Not compatible |
| Error ↔ trace linking | ✅ Automatic | ✅ Automatic (via `otlpIntegration`) |
| Package size | Larger (includes OTel) | Smaller (light mode) |
---
### For Each Agreed Feature
Load the corresponding reference file and follow its steps:
| Feature | Reference file | Load when... |
|---------|---------------|-------------|
| Error Monitoring | `references/error-monitoring.md` | Always (baseline) — captures, scopes, enrichment, beforeSend |
| OTLP Integration | See [OTLP Integration](#otlp-integration-otel-first-projects--nodejs-only) above | OTel tracing detected — **replaces** native Tracing |
| Tracing | `references/tracing.md` | OTel auto-instrumentation, custom spans, distributed tracing, sampling; **skip if OTel tracing detected** |
| Logging | `references/logging.md` | Structured logs, `Sentry.logger.*`, log-to-trace correlation |
| Profiling | `references/profiling.md` | Node.js only — CPU profiling, Bun/Deno gaps documented; **skip if OTel tracing detected** |
| Metrics | `references/metrics.md` | Custom counters, gauges, distributions |
| Runtime Metrics | See inline below | Automatic memory, CPU, and event loop metrics for Node.js and Bun |
| Crons | `references/crons.md` | Scheduled job monitoring, node-cron, Bull, Agenda, Deno.cron |
| AI Monitoring | Load `sentry-setup-ai-monitoring` skill | OpenAI, Anthropic, LangChain, Vercel AI, Google GenAI |
For each feature: read the reference file, follow its steps exactly, and verify before moving on.
### Runtime Metrics
Automatically collect Node.js and Bun runtime health metrics (memory, CPU utilization, event loop delay/utilization, uptime) at a configurable interval. Metrics appear in Sentry's Metrics product under the `node.runtime.*` / `bun.runtime.*` namespace.
**Node.js** — add `nodeRuntimeMetricsIntegration()` to your `instrument.js`:
```javascript
const Sentry = require("@sentry/node");
Sentry.init({
dsn: process.env.SENTRY_DSN,
integrations: [
Sentry.nodeRuntimeMetricsIntegration(),
// Optional: change collection interval (default 30 000 ms)
// Sentry.nodeRuntimeMetricsIntegration({ collectionIntervalMs: 60_000 }),
],
});
```
Metrics collected by default: `node.runtime.mem.rss`, `node.runtime.mem.heap_used`, `node.runtime.mem.heap_total`, `node.runtime.cpu.utilization`, `node.runtime.event_loop.delay.p50`, `node.runtime.event_loop.delay.p99`, `node.runtime.event_loop.utilization`, `node.runtime.process.uptime`.
**Bun** — add `bunRuntimeMetricsIntegration()` to your `instrument.ts`:
```typescript
import * as Sentry from "@sentry/bun";
import { bunRuntimeMetricsIntegration } from "@sentry/bun";
Sentry.init({
dsn: process.env.SENTRY_DSN,
integrations: [
bunRuntimeMetricsIntegration(),
// Optional: change collection interval (default 30 000 ms)
// bunRuntimeMetricsIntegration({ collectionIntervalMs: 60_000 }),
],
});
```
Metrics collected: same as Node.js except no event loop delay percentiles (unavailable in Bun). Prefixed with `bun.runtime.*`.
---
## Configuration Reference
### `Sentry.init()` Core Options
| Option | Type | Default | Notes |
|--------|------|---------|-------|
| `dsn` | `string` | — | Required. Also from `SENTRY_DSN` env var |
| `tracesSampleRate` | `number` | — | 0–1; required to enable tracing; **do not set when using OTLP path** |
| `dataCollection` | `object` | conservative unless set | Fine-grained control over auto-collected categories (`userInfo`, `cookies`, `httpHeaders`, `httpBodies`, `queryParams`, `genAI`). When omitted, the SDK falls back to `sendDefaultPii` (default `false`). Passing the object — even `{}` — flips unset categories to their permissive defaults; opt out per category. |
| `includeLocalVariables` | `boolean` | `false` | Add local variable values to stack frames (Node.js) |
| `enableLogs` | `boolean` | `false` | Enable Sentry Logs product (v9.41.0+) |
| `environment` | `string` | `"production"` | Also from `SENTRY_ENVIRONMENT` env var |
| `release` | `string` | — | Also from `SENTRY_RELEASE` env var |
| `debug` | `boolean` | `false` | Log SDK activity to console |
| `enabled` | `boolean` | `true` | Set `false` in tests to disable sending |
| `sampleRate` | `number` | `1.0` | Fraction of error events to send (0–1) |
| `shutdownTimeout` | `number` | `2000` | Milliseconds to flush events before process exit |
### `nativeNodeFetchIntegration()` Options
Configures outgoing `fetch`/`undici` span capture. Since `@opentelemetry/[email protected]`, response headers like `content-length` are no longer captured automatically — use `headersToSpanAttributes` to opt in:
```javascript
Sentry.init({
integrations: [
Sentry.nativeNodeFetchIntegration({
headersToSpanAttributes: {
requestHeaders: ["x-request-id"],
responseHeaders: ["content-length", "content-type"],
},
}),
],
});
```
| Option | Type | Default | Notes |
|--------|------|---------|-------|
| `breadcrumbs` | `boolean` | `true` | Record breadcrumbs for outgoing fetch requests |
| `headersToSpanAttributes.requestHeaders` | `string[]` | — | Request header names to capture as span attributes |
| `headersToSpanAttributes.responseHeaders` | `string[]` | — | Response header names to capture as span attributes |
### `otlpIntegration()` Options (`@sentry/node-core/light/otlp`)
For OTel-first projects using `@sentry/node-core/light`. Import: `import { otlpIntegration } from '@sentry/node-core/light/otlp'`.
| Option | Type | Default | Purpose |
|--------|------|---------|---------|
| `setupOtlpTracesExporter` | `boolean` | `true` | Auto-configure OTLP exporter to send spans to Sentry; set `false` if you already export to your own Collector |
| `collectorUrl` | `string` | `undefined` | OTLP HTTP endpoint of an OTel Collector (e.g., `http://localhost:4318/v1/traces`); when set, spans are sent to the collector instead of the DSN-derived Sentry endpoint |
### Graceful Shutdown
Flush buffered events before process exit — important for short-lived scripts and serverless:
```javascript
process.on("SIGTERM", async () => {
await Sentry.close(2000); // flush with 2s timeout
process.exit(0);
});
```
### Environment Variables
| Variable | Purpose | Runtime |
|----------|---------|---------|
| `SENTRY_DSN` | DSN (alternative to hardcoding in `init()`) | All |
| `SENTRY_ENVIRONMENT` | Deployment environment | All |
| `SENTRY_RELEASE` | Release version string (auto-detected from git) | All |
| `SENTRY_AUTH_TOKEN` | Source map upload token | Build time |
| `SENTRY_ORG` | Org slug for source map upload | Build time |
| `SENTRY_PROJECT` | Project slug for source map upload | Build time |
| `NODE_OPTIONS` | Set `--import ./instrument.mjs` for ESM | Node.js |
### Source Maps (Node.js)
Readable stack traces in production require uploading source maps with `@sentry/cli` or the webpack/esbuild/rollup plugins — for example, an inject + upload step in your build:
```json
{
"scripts": {
"build": "tsc && sentry-cli sourcemaps inject ./dist && sentry-cli sourcemaps upload ./dist"
}
}
```
Upload needs a `SENTRY_AUTH_TOKEN` (a build-time secret). For creating the token and wiring it into CI, see [`sentry-source-maps`](../sentry-source-maps/SKILL.md).
---
## Verification
After setup, verify Sentry is receiving events:
```javascript
// Add temporarily to your entry file or a test route, then remove
import * as Sentry from "@sentry/node"; // or @sentry/bun / @sentry/deno
Sentry.captureException(new Error("Sentry test error — delete me"));
```
Or trigger an unhandled exception:
```javascript
// In a route handler or startup — will be captured automatically
throw new Error("Sentry test error — delete me");
```
Then check your [Sentry Issues dashboard](https://sentry.io/issues/) — the error should appear within ~30 seconds.
**Verification checklist:**
| Check | How |
|-------|-----|
| Error captured | Throw in a handler, verify in Sentry Issues |
| Tracing working | Check Performance tab — should show HTTP spans |
| `includeLocalVariables` working | Stack frame in Sentry should show variable values |
| Source maps working | Stack trace shows readable file names, not minified |
---
## Phase 4: Cross-Link
After completing backend setup, check for companion services:
```bash
# Frontend companion
ls frontend/ web/ client/ ui/ 2>/dev/null
cat package.json 2>/dev/null | grep -E...
© getsentry · Apache-2.0 · getsentry/sentry-for-ai