`compliance_scan`

Vision AI read

What it does

Scores a content item against an embedded rule library covering Meta Ads, TikTok, YouTube and LinkedIn. Returns per-platform compatibility, an overall risk bucket, a safe_for_ads boolean, and a list of cited violations.

Every violation references a stable rule key (e.g. meta.body.scale_closeup) so you can act on the output programmatically rather than parsing prose.

Input schema

Name	Type	Required	Default	Description
id	string	required	`—`	Content id.
platforms	array<string>	optional	`all 4`	Subset of `meta`, `tiktok`, `youtube`, `linkedin`.

Example call

json{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "compliance_scan",
    "arguments": {
        "id": "c_abc123",
        "platforms": ["meta", "tiktok"]
    }
  }
}

Response shape

json{
  "content_id": "c_abc123",
  "upload_id": "u_xyz789",
  "scanned_at": 1716301234,
  "model": "claude-sonnet-4-6",
  "overall_risk": "high",
  "safe_for_ads": false,
  "platform_compatibility": {
    "meta": "warn",
    "tiktok": "ok",
    "youtube": "skipped",
    "linkedin": "skipped"
  },
  "violations": [
    {
      "rule": "meta.brand.third_party_logo",
      "platform": "meta",
      "category": "brand_visibility",
      "frame_timestamp_estimate": "0:08",
      "severity": "high",
      "suggestion": "Blur the La Marzocco logo visible at 0:08.",
      "rule_text": "Third-party brand logos visible — must be blurred."
    }
  ],
  "recommendations": [ "Re-render with logo blur before Meta upload." ],
  "notes": "Espresso tutorial with prominent brand placement.",
  "evidence": { "delivery_url": "https://lumiqa.io/api/deliver/..." },
  "limitations": [ "Frame extraction is not available inside Cloudflare Workers..." ]
}

Usage from Claude Desktop

prompt"Is c_abc123 safe to run as a Meta ad? If not, list every violation with the rule key."

Usage from Python

pythonimport requests

LUMIQA = "https://lumiqa.io/mcp/your-slug"
HEADERS = {"Authorization": "Bearer lk_live_xxxxxxxxxxxx"}

r = requests.post(LUMIQA, headers=HEADERS, json={
    "jsonrpc": "2.0", "id": 1,
    "method": "tools/call",
    "params": {
        "name": "compliance_scan",
        "arguments": {"id": "c_abc123", "platforms": ["meta"]}
    },
})
print(r.json())

Usage from Node

javascriptconst r = await fetch("https://lumiqa.io/mcp/your-slug", {
  method: "POST",
  headers: {
    "Authorization": "Bearer lk_live_xxxxxxxxxxxx",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    jsonrpc: "2.0", id: 1,
    method: "tools/call",
    params: {
      name: "compliance_scan",
      arguments: { id: "c_abc123", platforms: ["meta"] },
    },
  }),
});
const data = await r.json();
console.log(data);

Required env vars / integrations

Pages secret ANTHROPIC_API_KEY.
Optional USE_PREMIUM_COMPLIANCE=true to switch to Opus for the audit pass.
R2 binding UPLOADS + MCP_DELIVERY_SECRET.

Known limitations

Frame-level scoring is heuristic today. Workers can't run FFmpeg, so purely visual rules are inferred from metadata + the signed URL handed to the model. Output is honest about this in limitations.
Violations whose rule key isn't in the policy library are dropped — the model can't invent constraints.
Severity ladder: low < medium < high < block. safe_for_ads is false as soon as any high or block is present.

compliance_scan