AgentPay MCP Server

Agent-to-Agent micropayments via x402 + MCP — let AI agents autonomously pay for API access with USDC on Base.

Install

Option A — Remote MCP（推荐，无需本地安装）

在 Cursor / Claude / 任意支持 Streamable HTTP 的客户端添加：

https://agentpay-xhs-production.up.railway.app/mcp

官方 Registry 条目：io.github.yyxqqq777/agentpay

Option B — 本地 stdio（uvx / PyPI）

{
  "mcpServers": {
    "agentpay": {
      "command": "uvx",
      "args": ["mcp-server-agentpay"],
      "env": {
        "AGENT_PRIVATE_KEY": "0xYourAgentPrivateKey"
      }
    }
  }
}

只需填付款钱包私钥（需有 USDC）。网关地址已内置，一般不用改。

或永久安装：

pip install mcp-server-agentpay
# or
uv tool install mcp-server-agentpay

Tools

Example prompts

• "帮我看看这条小红书笔记写了什么：https://www.xiaohongshu.com/explore/..."
• "拉一下这个博主最近发的笔记，user_id=..."
• "Search wholesale prices for wireless earbuds under $3"
• "What's my AgentPay wallet and spending limit?"

Environment Variables

How it works

MCP Client  →  mcp-server-agentpay  →  x402 Gateway  →  Paid API
              (auto 402 → sign → retry)   (verify USDC)

1. Tool call hits a paid API endpoint
2. Gateway returns 402 Payment Required with USDC payment details
3. MCP server signs payment and retries automatically
4. JSON data flows back to the LLM

Self-host the Gateway

If you operate the paid API, deploy the gateway separately:

pip install "mcp-server-agentpay[gateway]"
# .env: PAYMENT_MODE=production + upstream keys + your receive wallet
agentpay-gateway

Real USDC settlement (x402)

Testnet (Base Sepolia) — use https://x402.org/facilitator, network eip155:84532.

# Fund a test agent with Sepolia USDC, then:
PYTHONPATH=src python scripts/test_x402_settlement.py

Mainnet — set PAYMENT_NETWORK=eip155:8453 and a mainnet facilitator (https://api.cdp.coinbase.com/platform/v2/x402 or PayAI). USDC goes to WALLET_RECIPIENT_ADDRESS.

See .env.example.

Publish to MCP Registry

This package is designed for the official MCP Registry.

# 1. Publish to PyPI
pip install build twine
python -m build && twine upload dist/*

# 2. Publish to MCP Registry (packages + remotes)
./scripts/publish_all.sh

Or create a GitHub Release (v0.1.0) — Actions uses Trusted Publishing to PyPI then updates the Registry.

Before publishing, update server.json and README mcp-name with your GitHub namespace (io.github.<username>/agentpay).

Development

git clone https://github.com/yyxqqq777/mcp-server-agentpay
cd mcp-server-agentpay
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
pytest tests/ -v

License

MIT