MCP Tool Reference

Around 80 tools across nine groups. Every tool carries MCP annotations, so a host can tell read-only from destructive before invoking anything.

Detection quality matches the REST API exactly. Both surfaces are backed by the same tool definitions from the @tuteliq/mcp package.

Detection

Tool	Purpose
`detect_grooming`	Predatory patterns across an adult-to-minor conversation
`detect_bullying`	Harassment, exclusion, intimidation, gaming toxicity
`detect_unsafe`	Self-harm, violence, hate, threats, sexual content
`analyze`	Quick combined safety check
`analyse_multi`	Up to 10 detectors over one text in a single call
`batch_analyze`	Up to 50 items in one request, for bulk triage
`analyze_emotions`	Emotional state indicators across a conversation

Grooming is scoped to adult-to-minor dynamics. If every participant is a known adult the endpoint short-circuits and says so; use coercive control or social engineering instead.

Fraud and exploitation

Tool	Purpose
`detect_social_engineering`	Pretexting, urgency fabrication, authority impersonation
`detect_app_fraud`	Fake investment platforms, phishing apps, subscription traps
`detect_romance_scam`	Love-bombing, financial requests, identity deception
`detect_mule_recruitment`	Money mule recruitment and account-sharing offers
`detect_gambling_harm`	Chasing losses, concealment, underage gambling
`detect_coercive_control`	Isolation, financial control, monitoring, threats
`detect_vulnerability_exploitation`	Targeting of elderly, disabled or financially distressed people
`detect_radicalisation`	Extremist rhetoric, us-versus-them framing, ideological grooming
`detect_tfgbv`	Technology-facilitated gender-based violence
`detect_distress_signals`	Acute distress indicators
`detect_emotional_distress`	Sustained emotional-harm patterns

Media

Tool	Purpose
`analyze_image`	Visual safety analysis plus OCR text extraction
`analyze_voice`	Transcription, then safety analysis of the transcript
`analyze_video`	Key-frame extraction and analysis (mp4, mov, avi, webm, mkv)
`analyze_document`	Per-page multi-endpoint PDF analysis with chain-of-custody hashing

Media costs more than text: an image is 7 credits, a voice clip 21, a video 95. See credits.

Synthetic content

detect_synthetic_text, detect_synthetic_image, detect_synthetic_audio and detect_synthetic_video identify AI-generated media in each modality.

Incidents and moderation

Tool	Purpose
`get_incidents_overview`	Queue snapshot with breakdowns by category, severity, source, status
`list_incidents`	Filterable queue, each row carrying `recommended_actions`
`moderation_queue`	Prioritised working set in one call
`get_incident`	Full record for one incident
`get_incident_trends`	Counts bucketed by hour, day or week
`review_incident`	Record a moderator decision
`batch_review_incidents`	One decision across up to 100 incidents

Agents are limited to confirm on the review tools. See moderation workflows.

Guidance and reporting

get_action_plan produces age-appropriate guidance for responding to a situation. generate_report produces an incident report suitable for professional review.

Governance and compliance

Tool	Purpose
`get_audit_receipt`	Signed EU AI Act Article 12 receipt for a specific request
`get_audit_logs`	Account-level action history
`register_encryption_key`	Register or rotate an RSA public key for end-to-end encryption
`get_encryption_key`	Fetch the currently registered key
`revoke_encryption_key`	Revoke it; new incidents fall back to server-side encryption
`export_account_data`	GDPR Article 20, portability
`rectify_data`	GDPR Article 16, correction
`delete_account_data`	GDPR Article 17, erasure
`record_consent`, `get_consent_status`, `withdraw_consent`	Consent ledger
`log_breach`, `list_breaches`, `get_breach`, `update_breach_status`	Breach register for Article 33

Set expectations correctly on erasure: Tuteliq stores no user content in the first place. An erasure request removes the abstracted detection records, and cannot remove message text because none was ever kept. That is usually a stronger answer than a completed deletion, and worth stating plainly.

Policy and automation

get_policy and set_policy manage account-wide thresholds. list_policy_rules, create_policy_rule, get_policy_rule, update_policy_rule, delete_policy_rule and evaluate_policy_rules drive rule-based automation. get_detection_settings, update_detection_settings and reset_detection_settings tune detector behaviour.

Threat intelligence

get_intelligence_trends, get_emerging_threats, get_risk_trends and get_weekly_digest surface cross-account, k-anonymised threat signals.

Verification, usage and webhooks

create_verification_session, get_verification_session and cancel_verification_session drive age and identity verification. get_usage_history, get_usage_by_tool, get_usage_monthly, get_usage_summary, get_usage_quota, get_pricing and get_pricing_details cover consumption and billing. list_webhooks, create_webhook, update_webhook, delete_webhook, test_webhook and regenerate_webhook_secret manage event delivery.

Tools that change or delete data

Most tools are read-only. These are annotated destructiveHint, so a host can prompt before invoking them. Worth reviewing before granting an agent access to a production account.

Tool	Effect
`delete_account_data`	Erases account data under GDPR Article 17. Not recoverable
`revoke_encryption_key`	New incidents fall back to server-side encryption; existing records stay readable only with the prior key
`rectify_data`	Amends stored records
`withdraw_consent`	Amends the consent ledger
`set_policy`	Changes detection thresholds account-wide
`update_detection_settings`, `reset_detection_settings`	Change detector behaviour account-wide
`update_policy_rule`, `delete_policy_rule`	Change or remove automation rules
`review_incident`	Records a moderator decision
`batch_review_incidents`	Records one decision across up to 100 incidents
`update_webhook`, `delete_webhook`	Change or remove event delivery
`regenerate_webhook_secret`	Invalidates the existing secret; receivers must be updated
`cancel_verification_session`	Ends a verification in progress
`update_breach_status`	Amends the breach register

Common parameters

Detection tools share a common set:

Parameter	Type	Required	Description
`content`	string	Yes	Text to analyse
`context`	object	No	Ages, relationship, platform and language hints
`include_evidence`	boolean	No	Return supporting excerpts with flagged phrases and weights
`support_threshold`	string	No	Minimum severity to include crisis helplines. `critical` always includes them
`external_id`, `customer_id`	string	No	Your identifiers, echoed in the response
`incident_moderation_enabled`	boolean	No	Override incident logging for this call: `true` forces it, `false` suppresses it. Omit for your account default. See incident logging
`verdict_only`	boolean	No	Fast mode: verdict without the per-message breakdown

verdict_only is the right trade when screening a live stream: screen everything fast, then re-run only what flags in standard mode for the detail.

Getting Started

SDKs

MCP Server

Integrations

Verification

Advanced

Compliance

MCP Tool Reference

Detection

Fraud and exploitation

Media

Synthetic content

Incidents and moderation

Guidance and reporting

Governance and compliance

Policy and automation

Threat intelligence

Verification, usage and webhooks

Tools that change or delete data

Common parameters

​Detection

​Fraud and exploitation

​Media

​Synthetic content

​Incidents and moderation

​Guidance and reporting

​Governance and compliance

​Policy and automation

​Threat intelligence

​Verification, usage and webhooks

​Tools that change or delete data

​Common parameters

Detection

Fraud and exploitation

Media

Synthetic content

Incidents and moderation

Guidance and reporting

Governance and compliance

Policy and automation

Threat intelligence

Verification, usage and webhooks

Tools that change or delete data

Common parameters