The core philosophy of DCL Trust Oracle is "don't trust the agent, trust the proof." Autonomous AI agents now perform actions with real consequences—financial, legal, reputational—but most agents are black boxes: there is no record of what it decided, why it decided that way, or whether that decision was tampered with afterward. DCL Trust Oracle is designed to fill this gap: every agent output is evaluated against policy in real time and sealed into a tamper-proof hash chain, forming a deterministic, cryptographically verifiable record of what happened and when. Any modification to a historical entry invalidates the entire chain. The validity of the record does not depend on trusting any party, including Fronesis Labs itself.
The system stores only cryptographic hashes and decision metadata, never raw content, thereby supporting verifiable, post-hoc forensic analysis across distributed AI agents.
Integration Methods
DCL Trust Oracle offers two ways to use it:
- REST API (webhook_server.py)—direct HTTP integration.
- MCP Server (mcp_server.py)—native Model Context Protocol integration for AI agents. The online address is https://mcp.fronesislabs.com/mcp（streamable-http）。该服务同时列于 Smithery (the remote URL points to the same production endpoint, see smithery.yaml) and the official MCP Registry.
Both servers share the same evaluation logic and tamper-proof chain (dcl_core.py), with identical pricing.
Running locally: the server runs at http://localhost:8080。 Production (hosted): Streamable HTTP transport—any MCP client can point directly to this address with no configuration required. Local development: the server runs at http://localhost:8081（streamable-http transport).
Tools and Endpoints
- Pre-Action Evaluation: intercept bad outputs before they reach users, wallets, or downstream systems.
- Session Management.
- Post-Action Forensics: when something does go wrong, precisely reconstruct what happened.
- Post-Action Scans.
DCL Update Sentinel (Continuous Skill Monitoring)
Continuous security auditing and release verification for AI Agent Skills and GitHub repositories. Sentinel monitors code updates via GitHub Webhooks, blocks regressions relative to a secure baseline, and automatically restores skill status after a clean rescan. Its features include:
- Idempotent Webhook Processing: built-in protection against duplicate GitHub delivery retries (X-GitHub-Delivery tracking).
- Automatic Recovery: skills that were blocked due to a security regression (status: blocked) are automatically restored to active after a clean, policy-compliant release.
- Strict Owner Verification: subscription renewal requires a cryptographically signed match with the registered owner wallet (owner_payer_ref).
- Baseline Policy Gate: registration requires an initial audit to pass (score >= 0.80). Compromised or failing repositories are immediately rejected (HTTP 422) to prevent invalid authorization setup.
Cryptography and Transaction Compliance (MCP only)
These tools are available only on the online MCP server (there are no corresponding REST routes in webhook_server.py).
Utilities (free, REST only)
Example Response
Verify the Hash Chain Yourself
You do not need to take the server's word for it. tx_hash is recalculated from the record's own fields, not merely linked to the previous row—so anyone can independently confirm that a record has not been edited after the fact, without calling back to this server. For a free, offline verification library, see @fronesis-labs/dcl-sdk (TS/JS) or dcl-core (Python).
Metering and Settlement
Each paid call described above is automatically metered and settled via the x402 protocol (USDC on Base). Two payment modes are supported:
- Per-call micropayments ($0.01–0.50): usefor real-time, on-demand evaluation.
- 30-day recurring license ($49.00): for continuous skill monitoring via DCL Update Sentinel.
This design eliminates API key configuration and invoicing overhead. Per-call pricing makes large-scale agent operations feasible (an autonomous system can initiate thousands of evaluation calls per day).
The REST API is x402-gated via fastapi-x402; the MCP server uses paymcp's Mode.X402, automatically paying for x402-capable clients and falling back to a guided payment link for clients without a configured wallet. Both settle to the same wallet, and neither has a bypass path—unpaid calls receive no verdict.
License: Apache License 2.0—see LICENSE.
