Dcl Trust Oracle MCP server

What it does

DCL Trust Oracle evaluates AI agent outputs against policy in real time and records verdicts in a tamper-evident hash chain that cannot be edited without invalidating the entire record. It serves developers building autonomous AI systems that need cryptographic proof of what decisions were made, when, and whether they were altered.

• Screen LLM outputs for policy violations, jailbreak attempts, and safety issues before they reach users or execute transactions.
• Audit past agent decisions by retrieving the tamper-evident record and forensic metadata for a given transaction hash.
• Monitor continuous drift in multi-turn agent sessions by initializing audit pipelines that track outputs over time.

Tools this server exposes

Tool	Does
dcl_evaluate_fast	Fast policy check for low-risk outputs, returns tamper-evident tx_hash.
dcl_evaluate_strict	Deep policy analysis for high-stakes outputs with higher confidence thresholds.
dcl_evaluate_jailbreak	Detects prompt injection patterns and role-hijacking attempts in agent outputs.
dcl_evaluate_safety	Baseline screening for known harmful text patterns optimized for high throughput.
dcl_evaluate_quality	Evaluates format adherence and contextual drift in agent outputs.
dcl_evaluate_batch	Bulk processing of up to 200 items per call for multi-turn history.
dcl_pipeline_start	Initializes a long-running audit session for continuous drift tracking.
dcl_audit_decode	Retrieves verdict, confidence, agent ID, and reason for a given transaction hash.
dcl_audit_decode_deep	Performs deep forensic audit including drift context, tamper-evidence indices, and environmental metadata.
dcl_evaluate_secrets	Scans for secret and credential leaks in agent outputs.
dcl_evaluate_pii	Detects personally identifiable information in agent outputs.
dcl_evaluate_jailbreak_crypto	Crypto-specific jailbreak and injection detection.
dcl_evaluate_wallet	Detects wallet secrets and sensitive cryptographic material in outputs.
dcl_evaluate_trade	Verifies trade decisions against compliance policy.
dcl_evaluate_mev	Screens for MEV and market-abuse compliance violations.
dcl_evaluate_signal	Screens for market signal fabrication in agent outputs.
dcl_evaluate_output_sanitizer	Final gate that sanitizes and verifies agent outputs before use.
dcl_commit	Seals a decision to the audit chain via Leibniz Layer crypto commit.

Before you enable it

An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.