证据追踪服务器

# Install
pip install mcp-server-tibet

# Add to Claude CLI
claude mcp add tibet -- python -m mcp_server_tibet

# Verify it works
claude mcp list
# tibet: ✓ Connected

Before TIBET:
  AI: "Loan approved" → Black box 🤷

After TIBET:
  AI: "Loan approved" → Full trail:
      WHO: loan_ai_v2
      WHAT: Approved application #4521
      WHEN: 2024-12-20T14:23:11Z
      WHY: "Customer meets all criteria"
      BASED ON: credit_score.pdf, income_verify.json
      SIGNATURE: 66360016ae08952e... ✓
      TRUST SCORE: 0.87 (HIGH)

# AI makes a decision - log it with TIBET
tibet_create_token(
    type="ai_decision",
    erin="Recommended treatment plan A for patient",
    eraan=["lab_results.json", "medical_history.pdf"],
    eromheen={
        "model": "medical-ai-v3",
        "confidence": 0.92,
        "hospital": "Amsterdam UMC"
    },
    erachter="Treatment A has 92% success rate for this condition",
    actor="medical_ai"
)
# → Token created with HMAC signature

# Later: Did anyone tamper with this decision?
tibet_verify_token(token_id="abc-123")

# Response:
{
    "valid": true,
    "integrity": "VERIFIED ✓",
    "trust_score": 0.89,
    "actor": "medical_ai",
    "state": "CREATED"
}

# Auditor asks: "Show me everything about this decision"
tibet_get_chain(token_id="abc-123")

# Response: Complete provenance from origin to now
{
    "chain_length": 3,
    "provenance": [
        {"actor": "medical_ai", "type": "ai_decision", ...},
        {"actor": "doctor_review", "type": "human_approval", ...},
        {"actor": "system", "type": "executed", ...}
    ]
}

# How trustworthy is this AI actor?
tibet_get_trust(actor="medical_ai")

# Response:
{
    "actor": "medical_ai",
    "trust_score": 0.89,
    "trust_level": "HIGH TRUST",
    "message": "FIR/A Trust Engine: medical_ai has HIGH TRUST (0.89)"
}

CREATED → DETECTED → CLASSIFIED → MITIGATED → RESOLVED