← 返回 Skills 市场
up2itnow

AgentPay MCP

作者 up2itnow · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ⚠ suspicious
299
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install agentpay-mcp
功能描述
MCP server that gives AI agents the ability to make payments, manage budgets, and handle billing -- directly from Claude Desktop, Cursor, Cline, or any MCP-c...
使用说明 (SKILL.md)

AgentPay MCP

Payment layer for AI agents via Model Context Protocol. AgentPay MCP gives your AI agent the ability to make payments, track spending, and manage payment channels -- without leaving the agent runtime.

Works with Claude Desktop, Cursor, Cline, Windsurf, and any MCP-compatible environment.

What It Does

  • Make payments: Send USDC on Polygon directly from agent tool calls
  • Budget management: Hard spending limits enforced per-session and per-day
  • Payment channels: Open Handshake58-style channels for sub-cent micropayments
  • CCTP bridge: Cross-chain USDC settlement (Ethereum, Base, Polygon, Arbitrum)
  • Non-custodial: Agent holds its own keys -- no third-party custody

Installation

npm install -g agentpay-mcp

Add to your MCP config (~/.config/claude/mcp.json or equivalent):

{
  "mcpServers": {
    "agentpay": {
      "command": "agentpay-mcp",
      "env": {
        "AGENT_PRIVATE_KEY": "0x...",
        "RPC_URL": "https://polygon-rpc.com",
        "MAX_TX_USDC": "25",
        "MAX_DAILY_USDC": "500"
      }
    }
  }
}

Tools Exposed

Tool Description
pay Send USDC to an address (enforces spend limits)
check_balance Query agent wallet balance
get_spending Current session and daily spend totals
open_channel Open a micropayment channel (Handshake58 compatible)
pay_channel Issue a signed payment voucher (zero gas)
close_channel Settle and reclaim unused channel funds
bridge_usdc Cross-chain USDC via CCTP

Usage Example

In Claude Desktop with AgentPay MCP configured:

"Pay 5 USDC to 0xABC...123 for the API call results"

The agent calls pay(to="0xABC...123", amount=5, token="USDC"). If within limits, it executes immediately. If over limits, it returns a request for human approval.

Environment Variables

Variable Required Description
AGENT_PRIVATE_KEY Yes Agent's signing key (0x-prefixed)
RPC_URL Yes Polygon JSON-RPC endpoint
MAX_TX_USDC No Per-transaction limit in USDC (default: 25)
MAX_DAILY_USDC No Daily spending limit in USDC (default: 500)

GitHub

https://github.com/AI-Agent-Economy/agentpay-mcp

License

MIT

安全使用建议
This skill gives an agent the ability to sign and send real cryptocurrency transactions. Before installing: (1) Do not provision your main wallet private key. Prefer a dedicated signing key or a constrained account funded only with a small test balance. (2) Require and verify human approval flows in your agent runtime for any payment above a tiny threshold. (3) Inspect the npm package and GitHub repository the SKILL.md references (pin a specific release/tag and review code) before running npm install -g. (4) Avoid storing private keys in plaintext config files; consider hardware signers or remote signing services that limit scope. (5) If you decide to proceed, set conservative MAX_TX_USDC and MAX_DAILY_USDC env values, run the package in an isolated environment, and audit logs for unexpected activity. Because this skill bundle contains no code to review, you cannot verify what the installed package will do — that uncertainty is the main reason this is flagged as suspicious.
功能分析
Type: OpenClaw Skill Name: agentpay-mcp Version: 1.0.0 The skill facilitates automated USDC payments on the Polygon network and requires a raw AGENT_PRIVATE_KEY in environment variables, which is a high-risk security practice for AI agents. It exposes powerful tools for financial transactions such as 'pay', 'open_channel', and 'bridge_usdc' in SKILL.md. While these capabilities align with the stated purpose of a payment MCP, the requirement for sensitive credentials and the potential for unauthorized fund movement via the external 'agentpay-mcp' npm package makes this skill inherently risky.
能力评估
Purpose & Capability
The skill is a payment/MCP server and therefore legitimately needs a signing key (AGENT_PRIVATE_KEY), an RPC endpoint (RPC_URL), and a Node runtime; those requirements are consistent with the stated functionality. Minor inconsistency: the skill is instruction-only in the registry (no code supplied or install spec), yet the SKILL.md instructs users to npm install -g agentpay-mcp and to modify MCP config files.
Instruction Scope
SKILL.md instructs storing the agent private key and RPC URL in MCP config or environment variables and adding a global npm binary to MCP server config (~/.config/claude/mcp.json). That places long‑lived signing material in a file/env accessible to the agent runtime and any processes that read that config. Because the skill describes autonomous payments, those instructions expand the agent's ability to create and sign transactions without additional checks unless external approvals are enforced.
Install Mechanism
The skill includes an npm install command and a GitHub URL in SKILL.md but provides no install spec or packaged code in the registry for review. There is no pinned release, checksum, or provenance provided in the skill bundle, so the actual code that would be installed cannot be audited from this skill entry alone.
Credentials
Requesting AGENT_PRIVATE_KEY and RPC_URL is proportionate for a payment tool, but AGENT_PRIVATE_KEY is extremely sensitive. The SKILL.md also recommends placing the key in MCP config or env vars — practices that increase exposure. No secondary mitigations (hardware signer, read-only/subkeys, or explicit human-approval gating) are mandated by the instructions.
Persistence & Privilege
The skill does not request always: true and does not claim to modify other skills, which is good. However, agent runtimes typically allow autonomous invocation by default; combined with the ability to sign transactions using a provided private key, autonomous invocation materially increases risk unless the agent runtime enforces explicit human approvals or very strict spend limits.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install agentpay-mcp
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /agentpay-mcp 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.0
Initial listing: Payment layer for AI agents via Model Context Protocol
元数据
Slug agentpay-mcp
版本 1.0.0
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 1
常见问题

AgentPay MCP 是什么?

MCP server that gives AI agents the ability to make payments, manage budgets, and handle billing -- directly from Claude Desktop, Cursor, Cline, or any MCP-c... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 299 次。

如何安装 AgentPay MCP?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install agentpay-mcp」即可一键安装,无需额外配置。

AgentPay MCP 是免费的吗?

是的,AgentPay MCP 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

AgentPay MCP 支持哪些平台?

AgentPay MCP 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 AgentPay MCP?

由 up2itnow(@up2itnow)开发并维护,当前版本 v1.0.0。

💬 留言讨论