MCP Portal
Sign in ← LTHN

ContentShield

eaas

Available

AI content detection and sycophancy scoring engine

About

Ethics as a Service scoring engine. Detects AI-generated content using
heuristic markers, grammar imprint analysis, and optional 1B/4B cascade
LEM inference. Tracks sycophancy across multi-turn conversations with
4-tier classification and differential analysis.

Backed by a Go binary with 25-dimensional atlas vectors (Qdrant),
session-level sycophancy profiling, and Layer B judge escalation
for uncertain cases.

Use when

  • • Scoring text for AI-generated content markers
  • • Analysing writing style with linguistic imprint fingerprinting
  • • Tracking sycophancy drift across a conversation
  • • Comparing prompt vs response grammar to detect parroting
  • • Evaluating content authenticity before publishing

Don't use when

  • • Plagiarism detection (this detects AI patterns, not copied text)
  • • Sentiment analysis (use a dedicated NLP tool)
  • • Content moderation or toxicity filtering
  • • Real-time chat filtering (latency-sensitive)

Connection

Call tools on this server via HTTP: 

curl -X POST https://mcp.lthn.ai/tools/call \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "server": "eaas",
    "tool": "score_content",
    "arguments": {}
  }'

Full setup guide →

Tools (9)

score_content

Score text for AI-generated content markers and sycophancy

Example prompts:

  • "Score this blog post for AI content markers"
  • "Is this response sycophantic?"
  • "Analyse this text for AI writing patterns"
Parameters
text * string

Text to score (minimum 20 characters)

prompt string

Original prompt that generated the text (enables sycophancy and differential analysis)

score_imprint

Generate a linguistic imprint fingerprint for text

Example prompts:

  • "Generate a writing fingerprint for this text"
  • "What's the linguistic profile of this content?"
Parameters
text * string

Text to analyse (minimum 20 characters)

session_start

Start a multi-turn sycophancy tracking session

Example prompts:

  • "Start tracking sycophancy for this conversation"
  • "Begin a scoring session"

session_turn

Score a conversation turn and track sycophancy trajectory

Example prompts:

  • "Score this turn in the conversation"
  • "Add this exchange to the session tracker"
Parameters
id * string

Session ID from session_start

text * string

AI response text to score (minimum 20 characters)

prompt string

User prompt that generated the response

session_get

Get the sycophancy profile for a scoring session

Example prompts:

  • "How sycophantic has this conversation been?"
  • "Show the session scoring profile"
Parameters
id * string

Session ID

score_model

Evaluate a model's semantic health from a prompt+response pair

Example prompts:

  • "How sovereign is this model's response?"
  • "Evaluate this prompt/response pair for alignment quality"
  • "Score this model output for ethical depth"
Parameters
prompt * string

The input prompt

response * string

The model's response (minimum 20 characters)

options object

Optional settings (judge, suites)

score_full

Run all scoring suites in one call

Example prompts:

  • "Run a full analysis on this text"
  • "Give me the complete scoring breakdown"
  • "Audit this response for compliance"
Parameters
prompt string

Original prompt (optional, enables semantic + delta scoring)

response * string

Text to analyse (minimum 20 characters)

options object

Optional settings (judge, suites)

score_retrieve

Retrieve a previously computed score by ID

Example prompts:

  • "Get the result for score ID sc_abc123"
  • "Look up a previous scoring result"
Parameters
id * string

Score ID (sc_, sm_, sf_, or si_ prefix)

score_health

Check if the scoring engine is running

Workflows (5)

Content Authenticity Check

Score a piece of content for AI markers

  1. score_content — Full heuristic + sycophancy + imprint analysis

Conversation Sycophancy Audit

Track sycophancy across a multi-turn conversation

  1. session_start — Create a tracking session
  2. session_turn — Score each prompt/response pair
  3. session_get — Review the sycophancy trajectory

Model Alignment Evaluation

Evaluate an AI model's output quality and sovereignty

  1. score_model — Semantic + delta + sycophancy analysis
  2. score_retrieve — Retrieve full results later by ID (cached 24h)

Full Compliance Audit

Run every scoring suite on a piece of content

  1. score_full — Returns heuristic + semantic + imprint + standard + composite

Writing Style Comparison

Compare linguistic fingerprints of two texts

  1. score_imprint — Generate imprint for text A
  2. score_imprint — Generate imprint for text B
  3. Compare the two imprint vectors — High similarity suggests same author/model
View Usage Analytics