# TrustGate / VERITAS — MCP trust checks, security audits and a public notary for AI agents (EN / ES / 中文) > [EN] Two separate businesses. SECURITY (private): free MCP-server trust checks, audits of contracts, skills, RAG, LLM outputs, GPU SLAs and backtests, and USDC escrow on Base. PUBLIC NOTARY (public by consent): sellers choose to be listed, rated 1-3 stars and reviewed; deals get a signed public receipt. Users never provide private keys. Policy: /privacy > [ES] Dos negocios separados. SEGURIDAD (privado): chequeos de confianza de servidores MCP, auditorías de contratos, skills, RAG, salidas de LLM, GPU y backtests, y escrow en USDC sobre Base. NOTARÍA PÚBLICA (pública por consentimiento): los vendedores eligen aparecer, con 1-3 estrellas y reseñas; los tratos reciben un acta pública firmada. Nunca pedimos claves privadas. Política: /privacy > [中文] 两项独立业务。安全(私密):免费的 MCP 服务器信任检查,对合约、技能、RAG、LLM 输出、GPU SLA 和回测的审计,以及 Base 链上的 USDC 托管。公共公证处(经同意后公开):卖家自愿登记,获得 1-3 星评级与评价;交易可获得签名的公开回执。用户从不需要提供私钥。政策:/privacy(中文仅为参考翻译,如有差异以英文版本为准) ## Verified On-Chain Identity / Identidad en Cadena / 链上验证身份 - **Base Mainnet Smart Contract (`TrustGateEscrow8183.sol`)**: `0xd4f77bf843b3c359b233aca46d7ff6604cbd0a5e` - **Sourcify v2 Full Source Verification**: `exact_match` (Match ID `#59283147`) — https://sourcify.dev/#/lookup/0xd4f77bf843b3c359b233aca46d7ff6604cbd0a5e - **Evaluator & Notary ECDSA Signer**: `0x861Ac49F8299C45Cca6bD9F0254080Fd82061Ac4` - **Liability Cap / Tope de Responsabilidad / 责任上限**: `STRICT_CAP_100PCT_OF_PROTOCOL_FEE` (1.0% - 1.5% fee only; Fail-Closed mode) ## Connection Endpoints / Puntos de Conexión / 接入端点 - **MCP Streamable HTTP (JSON-RPC 2.0)**: `POST http://own-adam.onrender.com/mcp` - **VERITAS-8183 Public Notary & 3-Star Seal Registry**: `GET http://own-adam.onrender.com/notary.json` - **Dual-Tier AI Catalog & x402 Micro-Bazaar**: `GET http://own-adam.onrender.com/catalog.json` - **MCP Auto-Discovery**: `GET http://own-adam.onrender.com/.well-known/mcp.json` - **ERC-8004 Agent Card**: `GET http://own-adam.onrender.com/.well-known/agent.json` - **TrustGate pricing descriptor**: `GET http://own-adam.onrender.com/.well-known/x402-bazaar.json` - **OpenAPI 3.1.0 Spec**: `GET http://own-adam.onrender.com/openapi.json` - **Free MCP Trust Badge**: `GET http://own-adam.onrender.com/badge.svg?server=` · record: `GET http://own-adam.onrender.com/trust?server=` - **Data & Privacy Policy**: `GET http://own-adam.onrender.com/privacy` (JSON) · `GET http://own-adam.onrender.com/privacy.md`. Security & escrow clients: PRIVATE. Registered sellers: PUBLIC by consent. Traffic: anonymous aggregates only. Users never provide private keys. ## START HERE (FREE): Is this MCP server safe to connect? 0. `check_mcp_server_trust` — Give any public MCP URL. Free, no signup, 5 checks/day: permissions, risky tools, prompt-injection signs and changes since last check. - **Full dossier (0.25 USDC on Base)**: preview the JSON for free with `get_trust_dossier_sample` or `GET /trust/sample.json`. - **Free skill review**: call `review_agent_skill_free` with empty arguments for a fictional JSON example, or with a root SKILL.md and related text files for findings by file and line. Sample: `GET /skills/review/sample.json`. ## 23 MCP Tools: Free Trust Check, 7 Audits, Escrow + Refund, Public Notary, Buyer Spending Limits & Opt-in Matchmaker 1. `discover_verified_stores` — ERC-8004 Dual-Tier Marketplace Filter (`INSTITUTIONAL_B2B` & `RETAIL_INDIE`) 2. `quote_multitoken_escrow` — Atomic Swap Quote (SOL, ETH, VIRTUAL, USDT -> Base USDC) 3. `open_escrow_job` — Lock USDC in ERC-8183 Vault 4. `submit_encrypted_deliverable` — [Vertical 1] Dataset Antimalware & 30-Row Blind Sampling 5. `audit_quant_backtest_escrow` — [Vertical 2] Quant Trading Backtest Auditor (Look-Ahead Bias, PnL, Sharpe, MaxDD) 6. `audit_smart_contract_escrow` — [Vertical 3] Solidity Smart Contract Compiler (`solc 0.8.38`) & Anti-Rugpull Scanner 7. `audit_api_sla_escrow` — [Vertical 4] x402 Paid API & Oracle SLA Verifier (Latency, Schema, Freshness) 8. `audit_agent_skill_escrow` — [Vertical 5] Agent Skill (`SKILL.md`), Prompt & MCP Plugin Anti-Trojan Scanner 9. `audit_rag_embeddings_escrow` — [Vertical 6] RAG Knowledge Base & Vector Embedding Anti-Poisoning Auditor 10. `audit_ai_inference_escrow` — [Vertical 7] Proof-of-Inference & Anti-Model-Downgrade Verifier 11. `audit_depin_gpu_sla_escrow` — [Vertical 8] DePIN & GPU Compute Benchmark SLA Auditor (FP16 TFLOPS, VRAM, Bandwidth) 12. `verify_onchain_escrow_status` — Verify Base Mainnet Contract, Sourcify & LEGION Ed25519 Proofs 13. `purchase_firstparty_catalog_item` — Instant purchase of TrustGate Official Curated Packages (from 1.00 USDC) 14. `register_and_certify_store` — Freemium ERC-8004 Store Onboarding with Mandatory Live Deterministic Audit (0 USDC upfront) 15. `inspect_veritas_8183_seal` — [FREE <5ms] VERITAS-8183 Bilateral Seal Inspector, 15-Provider 2★/3★ Leaderboard, 7-Question Continuous AI Trust Dossier, Version Drift (`v1 -> v2`) Detector & Pre-Execution Policy Firewall 16. `submit_veritas_tier_upgrade_exam` — VERITAS-8183 Seller & Smart Contract Tier Upgrade Exam (1 to 3 Stars with Proof-of-Reserve >=20% & Deterministic Code/Content Audit) 17. `manage_buyer_finops_spend_passport` — Buyer FinOps Spend Guard, Live Buyer Intent Registration, LLM Token Cost Optimizer, VIP Loyalty Discounts (up to 15% OFF) & 1-Time Single-Use Unlock Token Issuer 18. `notarize_and_witness_a2a_deal` — Millisecond A2A & Smart Contract Deal Witness + x402 Micro-Bazaar Executor (including 0.25 USDC Pay-Per-Sector Radar Intelligence `x402-radar-single-sector-intel`) 19. `match_verified_buyers_by_intent` — Exclusive 2-Star & 3-Star Seller Matchmaker to connect with verified buyers by available USDC budget and vertical ## Continuous AI Trust & 3-Layer Security Architecture (VERIFIED IN v0.7.0) - **Layer 1 (Static Security)**: 100% deterministic code, `solc 0.8.38` compilation, `SKILL.md`, and RAG vector verification. All 19 tools enforce explicit boolean annotations (`destructiveHint: false`, `openWorldHint: false`), MIT License, non-root Docker execution (`USER node`), and zero `process.env` secret reads. - **Layer 2 (Dynamic Pre-Execution Policy Firewall)**: Sub-2ms pre-flight action interceptor (`preExecutionCallCheck` in `inspect_veritas_8183_seal`) returning `ALLOW_EXECUTION` or `BLOCK_BEFORE_EXECUTION` before dangerous tool calls or over-budget FinOps actions run. - **Layer 3 (Historical Trust & Version Behavioral Drift `v1 -> v2 -> v3`)**: Tracks the 7 Continuous Trust Questions per agent (`q1_whatPermissionsDoesItHave`, `q2_whatToolsDoesItUtilize`, `q3_hasAttemptedUnauthorizedAccess`, `q4_areResultsReproducible`, `q5_hasBehaviorChangedAfterUpdate`, `q6_whatAuditsHasItPassed`, `q7_whatVulnerabilitiesRemainPending`).