For developers

Build moderation into anything.

Hush protects Instagram natively from the dashboard. But spam doesn't stop at Instagram — and not every workflow lives inside a SaaS dashboard. Our REST API, MCP server, and CLI let any platform, any internal tool, and any AI agent tap into the same detection engine.

Early access · EU-hosted · Free 5,000 comments to try

01 — REST API

Classify any text in a single call.

Send a comment, get a verdict. No SDK to install, no schema to learn — one POST request, one JSON response with verdict, score, and reasoning.

Request POST 
curl -X POST https://api.usehush.io/v1/classify \
  -H "Authorization: Bearer sk_..." \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Check my profile 😍🔥 link in bio",
    "platform": "instagram",
    "author": {
      "followers": 47,
      "following": 2,
      "posts": 3,
      "has_avatar": true
    }
  }'
Response 200 OK 
{
  "verdict": "SPAM",
  "score": 94,
  "category": "sexual_content",
  "explanation": "Sexual content lure with profile redirect; author signals consistent with throwaway account",
  "deciding_tier": "tier1_rules",
  "duration_ms": 12
}
  • Author context is optional — providing follower counts and avatar status improves accuracy on borderline cases
  • ~10–50ms typical response time — most comments resolve at Tier 1 (rules) without ever hitting the LLM
  • Multi-language — English, German, Spanish, Portuguese, French at launch
  • Webhook output — async hooks fire when spam is detected, so your system can react in real time

02 — MCP Server

Let your AI agents call Hush directly.

Hush ships with a Model Context Protocol server. Add it to Claude Desktop, Cursor, or any MCP-compatible agent, and your AI can classify comments as a native tool — no glue code, no wrapper service.

claude_desktop_config.json 
{
  "mcpServers": {
    "hush": {
      "url": "https://mcp.usehush.io/sse",
      "headers": {
        "Authorization": "Bearer sk_..."
      }
    }
  }
}
Then talk to your agent 
You:    I got these 6 comments on yesterday's launch post. Should I
        keep, hide, or reply?

        1. "Congrats!! Been waiting for this 🙌"
        2. "Just wired the team — when's the AMA?"
        3. "DM me @crypto_signals_pro for 10x returns"
        4. "Where can I read the changelog?"
        5. "🎁 You've been selected! Claim at bit.ly/free-iphone"
        6. "looks slow tbh, has anyone benchmarked it?"

Claude: Ran each through Hush. Here's what I'd do:

        Keep (3): #1 fan reaction, #2 genuine question,
                  #4 product enquiry.
        Hide (2): #3 Telegram crypto scam (94/100),
                  #5 fake giveaway with shortener (96/100).
        Reply  (1): #6 — critical but legitimate (Hush says
                    8/100, not spam). Worth a thoughtful reply.

        Want me to draft the reply to #6, or just confirm the hides?
  • Native tool calls — exposed as hush.classify, hush.classify_batch, hush.explain
  • Bring your own moderation logic — agents combine Hush's verdict with your own policy and take action via whatever other tools they have
  • Works with Claude, Cursor, Continue, OpenAI agents — anything that speaks MCP

03 — CLI

Pipe comments through Hush from your terminal.

Same engine, no service required. Useful for batch processing existing data, evaluating new detection rules, or integrating into CI/CD pipelines.

Install 
npm install -g @hush/cli
Classify a single comment 
hush classify "DM me on Telegram @crypto_profits"
Batch process a JSON file 
hush classify --file comments.json --output results.json
hush evaluate --file test-corpus.json   # accuracy report

What people build with this

Other platforms beyond Instagram

YouTube, TikTok, your own community forum, Discord, in-app chat. If you have user-generated text, Hush can classify it.

Custom moderation workflows

Combine Hush's verdict with your business rules. Auto-hide spam, flag uncertain cases for human review, route based on category — your logic, our detection.

AI agents that moderate

Build agents that triage comment sections, draft responses to legitimate users, and learn from your feedback. The MCP server makes Hush a first-class tool.

CI for content quality

Run Hush against test fixtures in CI. Catch regressions in your own content filters before they ship. The CLI is built for this.

Currently in early access.

The hosted API, MCP, and CLI are rolling out to early access users alongside the main Hush launch. If you want a key while we're still in the test phase, the waitlist is the same one for the product.

Join the waitlist Email us with use case