NuGet v3 Feed Extractor

Point at any NuGet v3 service index and export its packages as a clean, structured dataset — one row per package, straight from the feed's own self-describing v3 API.

The NuGet v3 protocol is self-describing: a feed's service index.json advertises its capabilities as a list of resources keyed by @type (SearchQueryService, RegistrationsBaseUrl, PackageBaseAddress, …). This actor resolves the feed's own advertised search endpoint and queries it, so one actor works against every v3 feed in existence — no per-host scraper, no hardcoded endpoints.

Unlike the nuget.org-only scrapers, this runner points at any v3 feed:

• nuget.org — the central public registry (450k+ packages)
• Azure Artifacts — Azure DevOps package feeds
• GitLab — the GitLab package registry
• MyGet, JFrog Artifactory, Sonatype Nexus
• self-hosted BaGet / Gitea / Forgejo package registries

Modes

The run mode is auto-selected from the input fields you set:

mode	when	emits
search	query is set	one row per package matching the search
list (default)	query is empty	one row per package the feed returns for an empty query
discovery	discoveryOnly: true	a single row describing the feed — advertised services, resource count, resolved search URL, total package count

Input

{
  "indexUrl": "https://api.nuget.org/v3/index.json",
  "query": "json",
  "prerelease": true,
  "maxRecords": 100
}

• indexUrl (required) — the v3 service index URL. Examples: https://api.nuget.org/v3/index.json, https://www.myget.org/F/nunit/api/v3/index.json.
• query — search term. Leave empty for list mode (all packages).
• prerelease — include beta/rc/alpha versions (bool).
• packageType — optional type filter (Dependency, DotnetTool, Template, …) — honoured by feeds that implement it.
• discoveryOnly — describe the feed only (one row).
• maxRecords — global cap; pagination stops when reached. Leave empty for everything.
• apiKey / extraHeaders — optional auth for private feeds (sent as X-NuGet-ApiKey / raw headers). Never required for public feeds, never logged.

Output

One flat, fully-nullable row per package:

field	example
packageId	Newtonsoft.Json
latestVersion	13.0.5-beta1
title / description / summary	Json.NET / …
authors / owners	James Newton-King
tags	["json"]
totalDownloads	8621110743
verified	true
versionCount	54
projectUrl / licenseUrl / iconUrl	…
packageTypes	["Dependency"]
latestVersionUrl / registrationUrl	…
_indexUrl / _searchUrl / _totalHits	feed + resolved-search metadata
_raw	the lossless original search entry

Every optional field is nullable and read by key presence, so a feed that spells totalDownloads differently (nuget.org totalDownloads vs MyGet totaldownloads) or omits iconUrl/tags never drops data or crashes — it just yields null for that column.

Behaviour

• A body that is not a v3 service index (HTML, 404 text, JSON without a resources list) fails fast with a clear message.
• A feed that advertises no SearchQueryService fails fast honestly in search/list mode (discovery mode still reports what it does support).
• An empty result is 0 records and a clean exit — never a crash.
• Transient 5xx / 429 responses are retried with backoff (honours Retry-After).

Pricing

Pay-per-event: one record event per package row returned.