ERC-8183 RAG Knowledge Base & Vector Embedding Anti-Poisoning Auditor & Escrow on Base.
Scored across 12 tools
The eight audit_* tools cleanly map to distinct verticals (skills, inference, API SLA, GPU, quant backtests, RAG vectors, Solidity, datasets), and the escrow lifecycle tools (discover/quote/open/submit/verify) are clearly separated. The one wrinkle is submit_encrypted_deliverable, which is really 'Vertical 1 dataset audit' but is named and framed as a submission/settlement action, so its boundary with audit_rag_embeddings_escrow and the audit family is slightly blurred.
All names are snake_case verb_noun and consistently readable (audit_*, discover_*, open_*, quote_*, submit_*, verify_*). The only deviation is that one audit operation is hidden under submit_ instead of the audit_ prefix used by its seven siblings, breaking the family pattern for that one tool.
Twelve tools is well within the ideal 3-15 band and each earns its place: eight narrow domain auditors plus five escrow lifecycle operations. No filler or near-duplicate tools.
The escrow flow covers discovery, quoting, locking funds, audited delivery with settlement/refund, and on-chain proof verification — a full happy path. However, there is no cancel/dispute/arbitrate operation, no list-my-jobs or balance/withdraw tool, so an agent hitting a stuck or contested escrow has a dead end.
[EN] Vertical 5 (Agent Skills & MCP Plugins): Scan SKILL.md, prompts & scripts against Trojan Skills (.env/PRIVATE_KEY exfiltration, eval/child_process, Base64 droppers & zero-width Unicode injection). | [ES] Línea 5 (Skills y Plugins MCP): Bloquea Skills Troyanos y robo de llaves privadas. | [中文] 垂直领域5 (AI 技能与 MCP 插件木马审计): 扫描 SKILL.md 与脚本，拦截私钥环境变量窃取、远程代码执行与隐形 Unicode 提示词注入。
Inputs: skillMarkdown (required), bundledScripts, requiredSections.
[EN] Vertical 7 (Proof-of-Inference & Anti-Model-Downgrade): Audit LLM output against token billing inflation, repetitive quantization collapse & constraint violations. | [ES] Línea 7 (Proof-of-Inference): Audita contra sobrefacturación de tokens y cambiazo de modelo IA. | [中文] 垂直领域7 (AI 推理防降级证明): 审计 Token 虚假超额计费、模型偷换导致的重复坍塌及指令约束合规性。
Inputs: responseText (required), billedOutputTokens (required), forbiddenTerms, mustIncludeTerms, +1 more.
[EN] Vertical 4 (x402 Paid API SLA Oracle): Verify paid API latency SLA (ms), mandatory JSON schema keys & cryptographic timestamp freshness. | [ES] Línea 4 (Oráculo SLA de APIs Pagas x402): Audita latencia, esquema JSON y frescura de datos. | [中文] 垂直领域4 (x402 付费 API 与预言机 SLA 审计): 验证响应延迟 (ms)、JSON 必填字段与时间戳新鲜度。
Inputs: apiResponsePayload (required), maxLatencyMs, requiredKeys, observedLatencyMs.
[EN] Vertical 8 (DePIN & GPU Compute SLA): Verify decentralized GPU hardware benchmarks (FP16 TFLOPS, available VRAM GB, memory bandwidth GB/s, thermal throttling & packet loss). | [ES] Línea 8 (Alquiler de GPUs y Cómputo DePIN): Verifica TFLOPS FP16, VRAM y temperatura. | [中文] 垂直领域8 (DePIN 去中心化 GPU 算力 SLA 审计): 验证 FP16 TFLOPS 算力、可用显存 (VRAM)、显存带宽与温度丢包率。
Inputs: benchmarkReport (required), minVramGb, minTflopsFp16, maxTempCelsius, +1 more.
[EN] Vertical 2 (Quant Backtests): Audit trading strategy backtests (NQ futures, crypto, forex) for Look-Ahead Bias, fake PnL accounting, Profit Factor, Max Drawdown & Sharpe Ratio. | [ES] Línea 2 (Backtests Cuantitativos): Audita historial de trading detectando Look-Ahead Bias y PnL falso. | [中文] 垂直领域2 (量化交易回测审计): 检测未来函数 (Look-Ahead Bias)、虚假盈亏 (PnL)、夏普比率、最大回撤与盈亏比。
Inputs: trades (required), minTrades, maxDrawdownPct, minProfitFactor.
[EN] Vertical 6 (RAG Knowledge Bases & Vector Embeddings): Verify exact vector dimensions, valid L2 norm (no zero/NaN vectors), zero cloned chunks & block RAG Data Poisoning. | [ES] Línea 6 (Memorias RAG y Vectores): Verifica dimensiones, norma L2 y bloquea envenenamiento RAG. | [中文] 垂直领域6 (RAG 向量知识库防投毒审计): 验证向量维度一致性、L2 范数有效性并拦截间接提示词注入投毒。
Inputs: chunks (required), minChunks, expectedDimension.
[EN] Vertical 3 (Solidity Smart Contracts): Compile Solidity source with real solc 0.8.38, scan against rugpull backdoors (selfdestruct, delegatecall, tx.origin, assembly sstore) & verify required ABI functions. | [ES] Línea 3 (Smart Contracts Solidity): Compila con solc real y bloquea puertas traseras. | [中文] 垂直领域3 (Solidity 智能合约安全审计): 真实 solc 编译、反跑路后门扫描 (selfdestruct/delegatecall/tx.origin) 与 ABI 接口验证。
Inputs: soliditySourceCode (required), contractName, requiredFunctions.
[EN] Query ERC-8004 AI store reputation registry and filter out low-trust sellers. | [ES] Consulta el registro ERC-8004 de tiendas IA y filtra por reputación mínima. | [中文] 查询 ERC-8004 AI 商店声誉注册表并过滤低信誉卖家。
Inputs: minReputation.
[EN] Lock USDC in non-custodial ERC-8183 Smart Contract on Base Mainnet to hire an AI store. | [ES] Bloquea fondos en USDC en el contrato ERC-8183 para contratar una Tienda IA. | [中文] 在 Base 主网 ERC-8183 非托管智能合约中锁定 USDC 资金以雇佣 AI 代理商店。
Inputs: storeId (required), tokenAmount (required), tokenSymbol (required), buyerId, +1 more.
[EN] Atomic multi-token swap quote (SOL, ETH, VIRTUAL, USDT -> Base USDC) with Dual-Lane Guardrail & volume discount (1% from 10 USDC). | [ES] Cotiza conversión atómica a USDC en Base con descuento por volumen. | [中文] 多代币原子兑换报价 (SOL/ETH/USDT 转 Base USDC) 及 ERC-8183 托管费用计算。
Inputs: tokenAmount (required), tokenSymbol (required).
[EN] Vertical 1 (Datasets): Audit encrypted CSV/JSON dataset (100% deterministic antimalware/clone scan + 30-row Vitter cryptographic blind sampling) & settle USDC or refund 100%. | [ES] Línea 1 (Datasets): Escáner Antimalware + Muestreo Ciego de 30 filas y liquidación automática. | [中文] 垂直领域1 (数据集审计): AES-256 加密 CSV/JSON 数据集反恶意软件与盲抽样审计，自动释放 USDC 或 100% 退款。
Inputs: jobId (required), rawDatasetJsonString (required).
[EN] Return Base Mainnet TrustGateEscrow8183 (0xd4f77bf843b3c359b233aca46d7ff6604cbd0a5e) cryptographic proofs, Sourcify exact_match verification & settled tx hashes. | [ES] Devuelve las pruebas criptográficas del contrato en Base Mainnet y Sourcify exact_match. | [中文] 获取 Base 主网托管合约 (0xd4f77bf8...0a5e) 的 Sourcify 源码精确匹配证明与链上交易哈希。
Inputs: no named parameters.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Payment-backed trust scores for AI agents on Base (ERC-8004). $0.01 per check via x402.
Tracea — legal identity (Know Your Agent) for AI agents, on-chain. ERC-8004 compatible.
Pay-per-call AI infra for agents: DeFi market data, LLM inference, STT/TTS. x402 v2, USDC on Base.
Agent credit issuance and scoring — programmable credit lines on Base L2
Add one secure layer between your agents and this server.
