@asuriq/mcp-server

Verify AI answers from inside Claude, Cursor, or any MCP-enabled client.

ASURIQ checks every AI response against 21 real-world databases — Semantic Scholar, PubMed, CrossRef, court records, FDA, WHO, and more. This MCP server gives you access to the full verification engine without leaving your conversation.

Install

npx @asuriq/mcp-server

Claude Desktop Setup

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "asuriq": {
      "command": "npx",
      "args": ["@asuriq/mcp-server"]
    }
  }
}

Tools

check_answer

Run the full ASURIQ analysis on any AI-generated text.

Parameters:

• text (required): The AI response to verify (min 20 chars)
• query (optional): The original question that was asked

Returns: Confidence score, hallucination risk, hedging analysis, entity verification, improvement hints, and more.

deliberate

Multi-model deliberation with voting and forensic audit trail.

Parameters:

• text (required): The AI response to deliberate on (min 20 chars)
• query (optional): The original question

Returns: Multiple model assessments (Defender/Attacker/Analyst), consensus voting, Shapley attribution, cognitive regime detection, and Fingerprint audit trail.

What it checks

• Confidence scoring — 7-axis decomposition (factual, completeness, recency, balance, domain, source, coherence)
• Hallucination detection — risk score with specific signals
• Entity verification — every study, statistic, and organization checked against 21 databases
• Hedging analysis — catches when AI sounds confident but qualifies every claim
• Code quality — deprecated APIs, security patterns, anti-patterns
• Gap analysis — what the AI should have said but didn't
• Improvement hints — better prompts generated from the weakest axis

Pricing

The MCP server uses the same ASURIQ engine. Free tier: 10 checks/day. Subscriptions and Deliberation Passes work across all surfaces.

Links

• Website: asuriq.dev
• API docs: asuriq.dev/api-docs
• All integrations: asuriq.dev/integrations

License

MIT