convez.ai
MCP ServerSegurançaIntermediário
ThoughtProof

ThoughtProof MCP

ThoughtProof/thoughtproof-mcp

Verificação adversarial de claims com consenso multi-modelo LLM

Ver no GitHub
InicialPluginScore · de 0 a 100

Servidor MCP para verificação adversarial de raciocínio em IA usando consenso multi-modelo. Claude, Grok e DeepSeek avaliam independentemente cada afirmação, retornando ALLOW, BLOCK ou UNCERTAIN com pontuação de confiança e atestação assinada via JWKS.

Instalação

$npx convez-ai ThoughtProof/thoughtproof-mcp

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 3 dias
Versão
fede-containment-fghij-2026-09-14
Issues abertas
5
Contribuidores
1

Casos de uso

  • Validar afirmações críticas antes de tomar decisões em código ou dados
  • Verificar saída de agentes IA em ambientes de alta importância (financeiro, médico, legal)
  • Detectar alucinações e raciocínio falho em tempo real durante execução de tarefas

Veredito

Ferramenta sólida para governança de IA em produção. O mecanismo adversarial multi-modelo reduz alucinações e oferece rastreabilidade criptográfica, mas o custo por verificação ($0.02–$0.08) limita uso em cenários de volume altíssimo.

README

A tradução para o português deste README ainda está na fila.

thoughtproof-mcp

npm version CI License: MIT

MCP server for ThoughtProof — verify AI reasoning with adversarial multi-model consensus.

3–4 LLMs (Grok, Gemini, DeepSeek, Sonnet) independently evaluate every claim. A dedicated red-team model critiques their verdicts. A synthesizer (Sonnet) weighs everything and returns ALLOW, BLOCK, or UNCERTAIN with confidence score and objections.

Quick Start

{
  "mcpServers": {
    "thoughtproof": {
      "command": "npx",
      "args": ["-y", "thoughtproof-mcp"],
      "env": {
        "THOUGHTPROOF_API_KEY": "tp_op_your_key_here"
      }
    }
  }
}

Works with Claude Desktop, Cursor, Windsurf, Cline, and any MCP-compatible client.

Tools

verify_claim

Verify any claim or AI-generated reasoning before acting on it.

ParameterTypeDefaultDescription
claimstring(required)The text to verify
stakeLevellow / medium / high / criticalmediumRisk level — higher stakes trigger deeper verification
domainfinancial / medical / legal / code / generalgeneralDomain context for specialized verification
speedfast / standard / deepstandardVerification depth

check_agent_score

Look up an agent's composite trust score on the ERC-8004 registry.

ParameterTypeDescription
agentIdstringAgent ID to look up
domainstringOptional domain filter

Example

In Claude Desktop or Cursor, just ask:

"Verify the claim: GPT-5 achieves 95% accuracy on MMLU-Pro"

The tool returns:

⚠️ UNCERTAIN (42% confidence)

Claim: "GPT-5 achieves 95% accuracy on MMLU-Pro"

Objections:
- Insufficient public benchmark data to confirm
- Historical accuracy claims have been overstated
- MMLU-Pro methodology has known ceiling effects

⚡ 3.2s | Adversarial Multi-Model Consensus

How It Works

Your AI Agent
    │
    ▼
┌──────────────────┐
│  thoughtproof-mcp │  ← MCP Server (this package)
└──────────────────┘
    │
    ▼
┌──────────────────┐
│  ThoughtProof API │  ← api.thoughtproof.ai (RV)
└──────────────────┘
    │
    ▼
┌───────────────────────────────────────────┐
│  Stage 1: Independent Evaluation       │
│  3–4 LLMs (Grok, Gemini, DeepSeek,     │
│  Sonnet) each examine the claim         │
│                                         │
│  Stage 2: Red-Team Critique             │
│  1 dedicated model challenges all       │
│  initial verdicts                        │
│                                         │
│  Stage 3: Synthesis                     │
│  Sonnet weighs verdicts + critique      │
│  → final decision                       │
└───────────────────────────────────────────┘
    │
    ▼
  ALLOW / BLOCK / UNCERTAIN
  + confidence % + objections

Pricing

SpeedModelsCost per verification
fast2$0.008
standard4$0.02
deep5+$0.08

Payment: API key (operator account) or x402 micropayment (USDC on Base).

API Key

Get an operator API key at thoughtproof.ai. Without a key, verifications use x402 micropayments automatically.

Configuration

Environment VariableDefaultDescription
THOUGHTPROOF_API_KEY(none)Operator API key
THOUGHTPROOF_BASE_URLhttps://api.thoughtproof.aiAPI base URL

Development

git clone https://github.com/ThoughtProof/thoughtproof-mcp.git
cd thoughtproof-mcp
npm install
npm run build
npm test
npm run dev          # Run with tsx (hot reload)
npm run inspect      # Test with MCP Inspector

Related

  • ThoughtProof — Decision verification for AI agents
  • pot-cli — CLI for reasoning verification
  • ERC-8004 — Autonomous Agent Registry

License

MIT — ThoughtProof

Ver o original no GitHub

Bom para

Relacionados