MCP Server Documentation

Overview

The Sentiqra MCP server exposes nine tools that any MCP-compatible AI client can call:

• detect_trust_safety - harmful language detection (16 languages)
• detect_cybersecurity - threat detection (15 languages)
• detect_crisis_intelligence - crisis signal detection (21 languages)
• detect_finance - market event detection (8 languages)
• detect_disinformation - manipulation technique detection (16 languages)
• detect_discrimination - discriminatory-language detection (16 languages)
• detect_emotions - emotion and sentiment detection (16 languages)
• detect_customer_service - support-intent detection (18 languages)
• detect_all - run all eight MCP-exposed analyzers in a single call (up to eight billable analyzer runs)

MCP tools expose the same response contract as POST /v1/analyze - no mode returns the complete taxonomy. Grouped tools return top-3 indicator groups and top-3 indicators per group; detect_emotions returns the top-3 emotion labels. Signal-match scores range from 0.0 to 1.0. Signal rows in the structured payload use the identical field shape as REST - analyzer, code, label, score, level (category/label), and parent on label-level rows - so one parser covers both surfaces. Results below the configured minimum signal score are filtered out automatically.

Execution & Cost Contract

The billing unit is the analyzer run - the same unit POST /v1/analyze and GET /v1/account/usage meter. One tool invocation is not one billable run: every tool declares the analyzers it may execute and its maximum billable runs, published in tools/list under _meta (contract sentiqra/mcp-tool/v1). Tool discovery is authoritative - this table is the deployed inventory it reports:

Transports

The MCP endpoint is reached over HTTPS - as a named server in AI clients, or directly from any HTTP client:

Claude Desktop Setup

Add Sentiqra to your Claude Desktop configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "sentiqra": {
      "url": "https://sentiqra.tech/v1/mcp",
      "headers": {
        "Authorization": "Bearer ${SENTIQRA_API_KEY}"
      }
    }
  }
}

Point the client at the hosted endpoint with your Bearer key - no local engine or data directory is required. After saving, restart Claude Desktop - you'll see the Sentiqra tools available in the tool palette. Dedicated (private) deployment is arranged through the controlled access path - contact Solutions Engineering.

Cursor Setup

Add Sentiqra to Cursor's MCP configuration:

~/.cursor/mcp.json

{
  "mcpServers": {
    "sentiqra": {
      "url": "https://sentiqra.tech/v1/mcp",
      "headers": {
        "Authorization": "Bearer ${SENTIQRA_API_KEY}"
      }
    }
  }
}

Point the client at the hosted endpoint with your Bearer key - no local engine or data directory is required. Dedicated (private) deployment is arranged through the controlled access path - contact Solutions Engineering.

HTTP Transport

The canonical MCP endpoint is POST /v1/mcp, aligned with the versioned REST prefix — POST /mcp remains as a compatibility alias; both are the same handler. Protocol versioning itself lives in the JSON-RPC protocolVersion field, not the URL path. The transport is stateless: every request is independent and initialize returns no Mcp-Session-Id — clients should not wait for or depend on a session header:

# Initialize
curl -X POST https://sentiqra.tech/v1/mcp \
  -H "Authorization: Bearer $SENTIQRA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","clientInfo":{"name":"my-app","version":"1.0"}}}'

# List tools
curl -X POST https://sentiqra.tech/v1/mcp \
  -H "Authorization: Bearer $SENTIQRA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'

# Call a tool
curl -X POST https://sentiqra.tech/v1/mcp \
  -H "Authorization: Bearer $SENTIQRA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"detect_trust_safety","arguments":{"text":"you are a worthless idiot","language":"en"}}}'

Tool Reference

Real Examples

These are live results from the deployed MCP endpoint, run under the analyzer's current deployed configuration - MCP tools do not take a configuration pin. Refresh the page and every response is identical - that's what deterministic means.

## Detection Results

**Language:** en

**Top Signals:**
1. Insults & Derogation (very high · 0.9)
   - Insult (very high · 0.9)
   - Belittling (very high · 0.9)

## Detection Results

**Language:** en

**Top Signals:**
1. Phishing Delivery (very high · 0.9)
   - Credential Lure (very high · 0.9)
   - Urgency Lure (very high · 0.9)

## Detection Results

**Language:** en

**Top Signals:**
1. Humanitarian Migration (very high · 0.9)
   - Humanitarian Access (very high · 0.9)
   - Displacement (very high · 0.9)

## Detection Results

**Language:** en

**Top Signals:**
1. Corporate Events (very high · 0.9)
   - Merger & Acquisition (very high · 0.9)

## Cross-Domain Detection Results

**Language:** en
**Analyzers:** trust-safety, cybersecurity, crisis-intelligence, finance, disinformation, discrimination, emotions, customer-service

**Trust & Safety:**
  1. Insults & Derogation (very high · 0.9)
     - Insult (very high · 0.9)
     - Belittling (very high · 0.9)

**Cybersecurity:**
  1. Phishing Delivery (very high · 0.9)
     - Credential Lure (very high · 0.9)
     - Urgency Lure (very high · 0.9)

**Crisis Intelligence:**
  1. Humanitarian Migration (very high · 0.9)
     - Humanitarian Access (very high · 0.9)
     - Displacement (very high · 0.9)

**Market Intelligence:**
  1. Corporate Events (very high · 0.9)
     - Merger & Acquisition (very high · 0.9)

**Disinformation:**
  1. Manipulation & Coordination (very high · 0.9)
     - Urgency Amplification (very high · 0.9)

**Discrimination:**
  1. Protected Class Bias (very high · 0.9)
     - Ethnic Exclusion (very high · 0.9)

**Emotion Analysis:**
  1. Negative Emotion (very high · 0.9)
     - Frustration (very high · 0.9)
     - Disappointment (very high · 0.9)

**Customer Service:**
  1. Support Intent (very high · 0.9)
     - Billing Issue (very high · 0.9)
     - Access Issue (very high · 0.9)

Protocol Details