NLP API MCP Server

Text analysis tools for MCP-enabled AI agents: toxicity detection, sentiment analysis, named entity recognition, PII detection, and language identification.

Tools

Tool	Description	Latency
analyze_toxicity	Score text for 6 harm categories (toxic, severe_toxic, obscene, threat, insult, identity_hate)	<15ms
analyze_sentiment	Classify positive/negative sentiment with confidence scores	<10ms
extract_entities	Extract named entities (PER, ORG, LOC, MISC) with span offsets	<50ms
detect_pii	Find PII (email, phone, SSN, credit card, IP, names) with optional redaction	<20ms
detect_language	Identify language from 176 supported (ISO 639-1 codes)	<1ms
check_nlp_service	Health check for NLP backends	<1s

Models

Capability	Model	Size
Toxicity	BERT-based classifier	~110M params
Sentiment	DistilBERT (general, financial, twitter variants)	~66M params
NER	BERT-NER (PER, ORG, LOC, MISC)	~110M params
PII	BERT-NER + regex ensemble	~110M params
Language	fastText (176 languages)	~917KB

How to Use

MCP Endpoint

https://<ACTOR_ID>.apify.actor/mcp?token=YOUR_APIFY_TOKEN

Example: Toxicity Detection

{
  "text": "You are a terrible person and I hope bad things happen to you"
}

Response:

{
  "toxic": 0.95,
  "severe_toxic": 0.12,
  "obscene": 0.08,
  "threat": 0.45,
  "insult": 0.89,
  "identity_hate": 0.03,
  "is_toxic": true
}

Example: PII Detection with Redaction

{
  "text": "Contact John Smith at john@example.com or 555-123-4567",
  "redact": true
}

Response:

{
  "pii_found": [
    {"text": "John Smith", "type": "PERSON", "start": 8, "end": 18, "score": 0.97},
    {"text": "john@example.com", "type": "EMAIL", "start": 22, "end": 38, "score": 1.0},
    {"text": "555-123-4567", "type": "PHONE", "start": 42, "end": 54, "score": 1.0}
  ],
  "count": 3,
  "redacted_text": "Contact [PERSON] at [EMAIL] or [PHONE]",
  "has_pii": true
}

Pricing

Tool	Price
Toxicity	$0.005/call
Sentiment	$0.005/call
NER	$0.005/call
PII Detection	$0.01/call
Language Detection	$0.002/call
Health Check	Free

Bulk discount: 50% off for >100K requests/month.

Features

• All models run on GPU for minimal latency
• GDPR/CCPA compliant PII detection with redaction
• 176 language support for language identification
• Domain-specific sentiment variants (general, financial, twitter)
• Structured JSON output compatible with any MCP client

Links

• API Docs: https://apim-ai-apis.azure-api.net
• Company: https://brainiall.com