Agentpay banner
garagon garagon

Agentpay

Security community

Description

Security plugin for Claude Code that protects agent-to-agent payments and transactions.

Installation

This entry records only its repository, not the path inside it, so there is no exact command to give. Open the source below and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

README

AgentPay

**Security plugin for Claude Code that protects agent-to-agent payments and transactions.**

Built at the [Kaszek & Anthropic Buenos Aires Hackathon 2026](https://dev.kaszek.com/) -- Open Track.

Install and use

As a Claude Code plugin:

claude plugins marketplace add garagon/agentpay
claude plugins install agentpay@agentpay

Or via npm:

npm i -g agentpay-security && agentpay install

Or from source:

git clone https://github.com/garagon/agentpay && cd agentpay
go build -o agentpay-bin . && ./agentpay-bin install

Once installed, AgentPay runs automatically on every tool call. Try the demo:

agentpay demo        # Run 7 attack scenarios
agentpay scan .      # Scan for compromised npm packages
agentpay audit       # View payment audit trail

What we built

AgentPay is a Claude Code plugin that intercepts financial tool calls in real time, detects when a compromised MCP server tampers with payments, and blocks fraud before money moves. It also detects and quarantines compromised npm packages (axios supply chain attack).

  • Real-time payment guard -- 5-stage security pipeline on every financial tool call
  • Payment drift detection -- Catches when a MCP changes the recipient, amount, or currency
  • Supply chain scanner -- Detects and removes compromised axios versions from node_modules
  • Tamper-evident audit -- Every decision logged with SHA-256 hash chain

No existing Claude Code plugin or MCP middleware provides financial security for agent transactions.

Why this matters

We are in the middle of a fundamental shift in how software is built and operated. The risks are compounding:

  1. 90% of code is AI-generated -- Agents write, review, and ship code with minimal human oversight. The attack surface is no longer just human error.
  2. MCP/library supply chain is vulnerable -- Compromised packages like axios (v1.7.8, 1.7.9) ship malicious postinstall scripts that exfiltrate c