API reference

One endpoint. Send text, get back IHRA-grounded antisemitism, anti-Israel bias, and context-omission signals with per-snippet attribution.

Base URL

https://magendetect.com

Authentication

Every request must include an x-api-key header. Create and rotate keys at /dashboard/api-keys after signing in.

x-api-key: mk_live_...

POST /v1/detect

Score a single piece of content. Synchronous — returns the full pipeline result.

Request body

field	type	required	description
`content`	string	yes	The text to analyze. Max 50,000 characters.
`title`	string	no	Article title. Improves triage accuracy.
`source`	string	no	Publisher name. Activates the trusted-source rule when applicable.
`language`	string	no	ISO code, e.g. `en` or `he`. Defaults to `en`.
`content_type`	string	no	Defaults to `article`. Free-form tag preserved in the audit record.
`metadata`	object	no	Arbitrary JSON persisted alongside the detection.

Example request

curl -X POST https://magendetect.com/v1/detect \
  -H "x-api-key: mk_live_..." \
  -H "Content-Type: application/json" \
  -d '{"title": "...", "source": "...", "content": "..."}'

Response

field	type	description
`id`	string	Unique detection ID (`det_...`).
`status`	string	`clear`, `review`, or `flagged`.
`scores`	object	Per-category max confidence.
`detections`	array	One entry per piece of evidence. See detection object below.
`triage`	object	Stage 1 triage: whether content needed deep analysis.
`verification`	object	Stage 3: accepted vs rejected signal count.
`calibration`	object	Stage 4: signal counts + `review_decision`.
`compliance`	object	DSA Article 34/15 flags.
`processing_ms`	number	Wall-clock pipeline time.
`audit_id`	string	Link to the persisted audit record.

Detection object

field	type	description
`category`	string	`antisemitism`, `anti_israel_bias`, or `context_omission`.
`subcategory`	string	Signal type, e.g. `classic_trope_conspiracy_control`.
`confidence`	number	0.0–1.0, after calibration.
`severity`	string	`low`, `medium`, `high`.
`attribution`	string	`publisher_voice` (1.0×), `quote` (0.6×), `rebuttal` (0.15×), `unclear` (0.5×).
`dog_whistle_hit`	bool	True if pre-scanner flagged this text.
`flagged_segments`	array	Character offsets into original content.
`definition`	object	Framework, reference (e.g. `IHRA-7`), description.
`rationale`	string	Model's explanation.
`recommended_action`	string	`block_or_review` or `editorial_review`.

Example response

{
  "id": "det_8f3a1c44",
  "status": "review",
  "processing_ms": 5439,
  "scores": { "antisemitism": 1.0, "anti_israel_bias": 0.05 },
  "detections": [{
    "category": "antisemitism",
    "subcategory": "classic_trope_conspiracy_control",
    "confidence": 1.0,
    "severity": "high",
    "attribution": "publisher_voice",
    "dog_whistle_hit": true,
    "definition": { "framework": "IHRA", "reference": "IHRA-7" },
    "rationale": "Classic antisemitic tropes about Jewish control."
  }],
  "calibration": { "review_decision": "human_review" }
}

Status legend

status	review_decision	meaning
`clear`	`auto_accept`	No actionable signals found.
`review`	`human_review`	Signals present but ambiguous. Attribution may soften a strong signal.
`flagged`	`auto_reject`	High-severity signal above threshold. Recommend review before publishing.

Rate limits

Each key has a monthly quota (default 10,000 for Starter). Resets automatically on the first of each month, UTC. When you hit the limit the API returns 429 rate_limit_exceeded.

Every authenticated response includes three headers:

header	value
`X-RateLimit-Limit`	Your quota for the current period.
`X-RateLimit-Remaining`	Calls remaining before cutoff.
`X-RateLimit-Reset`	Unix timestamp of the next reset.

Errors

code	error	when
400	`content_required`	Missing or empty `content`.
400	`invalid_json`	Body is not valid JSON.
401	`missing_api_key`	No `x-api-key` header.
401	`invalid_api_key`	Key does not exist or was revoked.
413	`content_too_large`	Exceeds 50,000 characters.
429	`rate_limit_exceeded`	Monthly quota consumed.
500	`detection_failed`	Pipeline did not complete.