Integrate Azure AI Content Safety API

Azure AI Content Safety REST SDK for TypeScript is a client for analyzing text and images for harmful content, with support for customizable blocklists. It is a REST client, not a class-based SDK - ContentSafetyClient is a function you call with the endpoint and a credential, returning the client instance.

It authenticates either with an API key via AzureKeyCredential or with DefaultAzureCredential from @azure/identity. Text analysis posts to /text:analyze with the text, the categories to check (Hate, Sexual, Violence, SelfHarm), and an output type - FourSeverityLevels (the default, returning 0, 2, 4, or 6) or EightSeverityLevels (returning 0-7) - and returns a severity score per category. Image analysis posts to /image:analyze with either base64-encoded content or a blob URL. Blocklist management covers creating a blocklist (a PATCH to /text/blocklists/{blocklistName}), adding items to it, listing blocklists, deleting one, and running text analysis against one or more named blocklists with a haltOnBlocklistHit option, returning any blocklist matches alongside the category severities.

The severity scale maps to a recommended action: 0 (Safe) means allow, 2 (Low) means review or allow with a warning, 4 (Medium) means block or require human review, and 6 (High) means block immediately. A helper pattern combines both checks: run /text:analyze with the blocklist names attached, then flag any category whose severity exceeds an allowed threshold and collect any blocklist term matches, returning a single isAllowed verdict alongside the flagged categories, the maximum severity seen, and the matched blocklist terms.

Use it when building content moderation for text or image user-generated content in a TypeScript/Node application, especially when you need custom blocklists layered on top of the built-in harm-category detection. Don't call it like a class constructor - new ContentSafetyClient(...) is wrong; it's a factory function, and every response must be checked with isUnexpected() before reading result.body, since the REST client doesn't throw automatically on API errors.

Input is text or an image (base64 content or a blob URL), plus the harm categories, output-severity type, and any blocklist names to check against. Output is a categoriesAnalysis array with a severity score per category, and, when blocklists are used, a blocklistsMatch array naming the matched terms and the blocklist they came from. Install it with:

npm install @azure-rest/ai-content-safety @azure/identity @azure/core-auth

It's the @azure-rest/ai-content-safety package paired with @azure/identity and @azure/core-auth, calling the Azure AI Content Safety REST API endpoints for text and image analysis (/text:analyze, /image:analyze) and blocklist CRUD (/text/blocklists/{blocklistName} and its item-management sub-paths).

TypeScript and Node.js developers building content moderation into an application - implementing the full blocklist lifecycle (create, add items, list items, remove items, delete) alongside category-severity checks - who want typed REST parameters like AnalyzeTextParameters and TextBlocklistItem rather than raw HTTP calls.