360 Secure SkillHub
360 安全技能中心

MnemoPay

面向自主 Agent 的信任与控制层:提供受治理的持久记忆、多种支付通道、预算与审批策略、身份与声誉、操作台账和审计证据,并以 MCP 工具组形式按权限分级暴露能力,默认仅开放基础工具组。

t49qnsx7qt-kpanks
t49qnsx7qt-kpanks
热度 49.74
最近更新 2026-06-09 04:32:56
分类:安全依赖:MCP 服务 / 第三方 API / 代码执行应用场景:财务 / 工作流

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add mnemopay
Metadata
name
mnemopay
description
Trust OS for AI agents that handle money or consequential actions. Adds governed memory, payments, reputation, identity, approvals, audit evidence, and durable browser/code/computer jobs.

MnemoPay

MnemoPay is the trust and control layer for autonomous agents. Use it when an agent needs persistent memory, payment rails, scoped authority, budgets, approvals, reputation, or verifiable evidence of what it did.

Install

npm install @mnemopay/sdk
claude mcp add -s user mnemopay -- npx -y @mnemopay/sdk mcp

Python:

pip install mnemopay

MCP Tool Groups

MnemoPay has a broad platform catalog, but exposes only the 14-tool essentials group by default.

npx -y @mnemopay/sdk mcp --tools=essentials
npx -y @mnemopay/sdk mcp --tools=governance,identity,skills
npx -y @mnemopay/sdk mcp --tools=agent_os

Never expose organization_admin, operator, or all to an untrusted agent. Those groups require explicit credentials and should be enabled only for authorized operators.

Capabilities

  • Persistent memory with decay, reinforcement, and Merkle integrity
  • Stripe, Paystack, Lightning, x402, MPP, and AP2 payment rails
  • Agent Reputation Scoring and portable KYA identity
  • Spend policies, capability tokens, approvals, and emergency stops
  • Action ledger, signed receipts, and Article 12 audit bundles
  • Governed skills and GridStamp spatial evidence
  • Durable hosted browser, coding, computer-use, skill, and brain jobs
  • Organization roles, members, invitations, policies, agents, and limits

Safety Rules

  • Require an explicit amount, currency, recipient, and reason before payments.
  • Prefer read-only tools until the user clearly requests an action.
  • Use approval workflows for high-risk, destructive, or high-value actions.
  • Never return API keys, invitation tokens, worker credentials, or secrets.
  • Keep admin/operator tools outside the default tool set.

Links