什么是 Proxy Pay MCP？

Proxy Pay MCP 是一款专门为 AI 代理提供安全金融层而设计的工具。通过集成此技能，开发者可以赋予其代理执行真实世界金融交易的能力，例如支付云服务费用、购买数字资产或管理订阅。该技能充当了高级代理逻辑与金融基础设施之间的桥梁，确保所有支出都受严格的、用户定义的策略约束。在 Openclaw Skills 生态系统中，Proxy Pay MCP 是构建需要在商业环境中独立运行的代理的关键组件。

此技能为代理提供了声明支付意图的能力，随后系统会根据支出策略对其进行评估。如果请求符合预设限制，将立即发行一张虚拟卡。这种架构确保了即使是完全自主的代理也处于财务控制之下，并具有包括商家限制和高额购买的人工审批阈值在内的多层保护。对于使用 Openclaw Skills 的开发者来说，这为代理化商业提供了一个生产级的解决方案。

下载入口:https://github.com/openclaw/skills/tree/main/skills/proxyhq/proxy-payments

安装与下载

1. ClawHub CLI

从源直接安装技能的最快方式。

2. 手动安装

将技能文件夹复制到以下位置之一

优先级：工作区 > 本地 > 内置

3. 提示词安装

将此提示词复制到 OpenClaw 即可自动安装。

Proxy Pay MCP 应用场景

• 管理和支付循环发生的 SaaS 订阅和 API 使用费。
• 为特定任务按需配置 AWS 或 Google Cloud 等云资源。
• 处理代理需要购买数字资产或研究材料的采购任务。
• 通过将交易与特定的代理意图匹配，实现自动化费用对账。
• 实施预算受限的研究项目，由 AI 代理管理其自身的资金分配。

1. AI 代理使用 proxy.intents.create 工具生成支付意图，并指定金额和商家。
2. 系统根据分配给代理的支出策略评估该意图，检查限制和约束条件。
3. 如果意图符合策略，则自动配置虚拟卡；否则，它将进入待定状态等待人工批准。
4. 一旦意图获批，代理将检索敏感的卡片详情（卡号、CVV）以完成结账流程。
5. 交易将被实时追踪，代理使用 proxy.receipts.attach 工具上传证据以供审计。
6. 所有金融活动都会针对初始意图进行对账，从而在 Openclaw Skills 中提供透明的审计追踪。

Proxy Pay MCP 配置指南

1. 在 useproxy.ai 注册 Proxy 账户并完成所需的 KYC/KYB 验证。
2. 在 Proxy 控制面板中创建一个“支出策略”和一个“代理”，以定义支出边界。
3. 通过运行以下命令使用 Claude Code 进行身份验证：

claude mcp add proxy --transport http https://mcp.useproxy.ai/api/mcp

4. 从 ~/.claude/.credentials.json 中提取您的访问令牌，并将其作为 PROXY_TOKEN 添加到您的环境变量中。
5. 更新您的 config/mcporter.json 文件以包含 Proxy MCP 服务器配置：

{
  "mcpServers": {
    "proxy": {
      "baseUrl": "https://mcp.useproxy.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer ${PROXY_TOKEN}"
      }
    }
  }
}

6. 通过调用 mcporter list proxy 确认设置，以确保所有工具均可用。

Proxy Pay MCP 数据架构与分类体系

Proxy Pay MCP 将数据组织成多个实体，以确保为 Openclaw Skills 用户提供清晰的财务报告和合规性。

Proxy Pay MCP

Give your OpenClaw agent a credit card. Proxy lets AI agents autonomously request and use virtual payment cards within policy-defined spending limits.

Features

• Intent-Based Payments — Agents declare what they want to buy, cards are auto-provisioned within policy
• Spending Policies — Set per-transaction limits, daily/monthly caps, merchant restrictions, and approval thresholds
• Virtual Cards — Single-use or limited cards issued instantly when policy allows
• Human-in-the-Loop — High-value intents require human approval before card issuance
• Balance & Funding — Check spending power, get ACH/wire/crypto deposit details
• Transaction Tracking — Monitor charges, attach receipts, reconcile against intents
• KYC/KYB Verification — Onboarding status and completion links

Setup

1. Create a Proxy Account

1. Go to useproxy.ai and sign up
2. Complete KYC/KYB verification
3. Create a Spend Policy with your desired limits
4. Create an Agent and assign the policy
5. Fund your balance (ACH, wire, or USDC on Solana)

2. Authenticate via OAuth

Proxy uses OAuth for MCP authentication. Use Claude Code to authorize, then extract the token for mcporter.

Step 1: Add Proxy MCP to Claude Code

claude mcp add proxy --transport http https://mcp.useproxy.ai/api/mcp

Step 2: Authorize in Claude Code

claude
# In session, run: /mcp
# Complete OAuth in browser

Step 3: Extract token

jq -r '.mcpOAuth | to_entries | .[] | select(.key | startswith("proxy")) | .value.accessToken' ~/.claude/.credentials.json

Step 4: Add to environment

# Add to ~/.clawdbot/.env
PROXY_TOKEN=eyJhbGciOiJ...

Step 5: Configure mcporter

Add to config/mcporter.json:

{
  "mcpServers": {
    "proxy": {
      "baseUrl": "https://mcp.useproxy.ai/api/mcp",
      "description": "Proxy Pay — AI agent payments",
      "headers": {
        "Authorization": "Bearer ${PROXY_TOKEN}"
      }
    }
  }
}

Step 6: Test

mcporter list proxy
mcporter call 'proxy.proxy.tools.list()'

Alternative: Agent Token (Autonomous)

For fully autonomous agents, use an agent token instead of OAuth:

1. In the Proxy dashboard, go to Agents and create an MCP token
2. Add to ~/.clawdbot/.env:

   PROXY_AGENT_TOKEN=proxy_agent_...
   

3. Configure mcporter with the agent token header:

   {
     "mcpServers": {
       "proxy": {
         "baseUrl": "https://mcp.useproxy.ai/api/mcp",
         "headers": {
           "Authorization": "Bearer ${PROXY_AGENT_TOKEN}"
         }
       }
     }
   }
   

Agent tokens are scoped to a single agent and its assigned policy.

Available Tools (25)

Onboarding & Profile

Balance & Funding

Policies

Payment Intents (Core Flow)

Cards

Transactions & Receipts

Merchant Lookup

Utilities

Usage Examples

Check Balance

mcporter call 'proxy.proxy.balance.get()'

Create a Payment Intent

mcporter call 'proxy.proxy.intents.create(
  description: "AWS monthly bill",
  merchantName: "Amazon Web Services",
  amount: 14999,
  currency: "USD"
)'

Amount is in cents (14999 = $149.99). If within policy limits, a virtual card is auto-issued.

Get Card Details for Payment

# Get intent with card info
mcporter call 'proxy.proxy.intents.get(intentId: "int_abc123")'

# Get sensitive card data (PAN, CVV) for checkout
mcporter call 'proxy.proxy.cards.get_sensitive(
  cardId: "card-uuid-here",
  intentId: "int_abc123",
  reason: "Paying AWS invoice"
)'

Simulate Before Spending

mcporter call 'proxy.proxy.policies.simulate(
  amount: 50000,
  currency: "USD",
  merchantName: "OpenAI",
  description: "API credits"
)'

Returns whether the intent would be auto-approved, need approval, or be denied.

Attach a Receipt

mcporter call 'proxy.proxy.receipts.attach(
  intentId: "int_abc123",
  url: "https://example.com/invoice-2026-01.pdf",
  description: "January invoice"
)'

Intent Flow

1. Agent calls proxy.intents.create with amount, merchant, and description
2. Policy evaluated — checks limits, merchant restrictions, approval thresholds
3. If auto-approved → virtual card issued instantly, returned in response
4. If needs approval → status = pending_approval, agent polls approval_status
5. Human approves/rejects via dashboard or intents.approve/intents.reject
6. Agent uses card — calls cards.get_sensitive for PAN/CVV to complete purchase
7. Transaction matched — webhook reconciles charge against intent
8. Receipt attached — agent uploads evidence for audit trail

Error Codes

Notes

• OAuth tokens give full account access (balance, funding, all agents' cards)
• Agent tokens are scoped to one agent and its policy — cannot see other agents' data
• Card-sensitive data requires both a cardId and intentId for compliance
• Amounts are in cents — $10.00 = 1000, $149.99 = 14999
• Policies are enforced server-side — agents cannot bypass limits
• Human approval tools (approve, reject) require OAuth (human) tokens

Resources

• Proxy Dashboard
• Proxy Documentation
• MCP Endpoint
• GitHub