MCP-Pay
Enables AI agents to discover and pay for MCP tools on a sub-cent, pay-per-call basis using x402 and Algorand USDC settlements.
README
MCP-Pay
An HTTP x402 monetization layer on Algorand that lets AI agents pay for MCP tools on a sub-cent, pay-per-call basis.
Screenshots
![]() |
![]() |
![]() |
![]() |
(All six captures live in docs/.)
Problem
Autonomous AI agents using the Model Context Protocol (MCP) cannot access paid tools: credit cards, API keys, and subscriptions are human workflows. Tool creators have no way to charge per call, and legacy chains are too slow or expensive for machine-to-machine micropayments.
Solution
MCP-Pay converts any MCP tool into an instantly monetized x402 endpoint on Algorand.
- An agent discovers tools at
GET /mcp/v1/tools(free, includes JSON schemas + prices). - The agent invokes
POST /mcp/v1/tools/:name/executewithout payment. - The server answers
402 Payment Requiredwith payment metadata (price in USDC, receiver address, network). - The agent's local middleware auto-signs an Algorand USDC transfer (ASA 31566704) and retries.
- The GoPlausible facilitator verifies and settles on-chain (<3.3s finality), the tool executes, and results return in one round-trip.
No accounts. No keys. No humans.
USP
| Typical paid API | MCP-Pay |
|---|---|
| One hardcoded endpoint | Tool registry: any tool becomes a priced service with a schema |
| Docs written for humans | Agent-native discovery: agents list, compare, and select tools programmatically |
| Subscriptions / prepaid credits | True pay-per-call settled on-chain per request |
| EVM gas kills micropayments | Algorand: ~0.001 ALGO fee, 3.3s finality |
Paid Tools Included
| Tool | What it does | Price |
|---|---|---|
web_research |
Reader-mode extraction of any public web page for LLM context | $0.002 |
market_data |
Live crypto spot prices from a keyless public feed | $0.001 |
contract_audit |
Static heuristic security scan of Solidity source | $0.005 |
Architecture
┌────────────────────────┐
│ AI AGENT (client/) │ 1. GET /mcp/v1/tools (discovery)
│ auto-paying MCP agent │ 2. POST .../execute → 402 challenge
└───────────┬────────────┘ 3. sign USDC txn locally
│ 4. retry + X-PAYMENT header
▼
┌────────────────────────┐
│ MCP-PAY SERVER (Hono) │ x402 paymentMiddleware protects each tool route
│ registry + tools/* │ executes tool only after verified settlement
└───────────┬────────────┘
▼
┌────────────────────────┐
│ GOPLAUSIBLE FACILITATOR│ /verify + /settle on Algorand TestNet
│ https://facilitator. │ USDC (ASA 31566704) transfer
│ goplausible.xyz │
└────────────────────────┘
Tech Stack
- TypeScript, Hono
@x402/core,@x402/avm,@x402/hono,@x402/fetch@x402-avm/extensions(Bazaar discovery)- GoPlausible facilitator, Algorand TestNet USDC
Run Locally
1. Prerequisites
- Node.js 18+, pnpm (
npm i -g pnpm) - Two funded Algorand TestNet accounts:
- Server wallet (receiver): address only
- Agent wallet (payer): mnemonic, kept local — never committed
Fund via Lora faucet, opt both into USDC in Lora, then get TestNet USDC from the Circle faucet (choose Algorand Testnet).
2. Configure
pnpm install
cp .env.example .env
Fill .env:
AVM_ADDRESS=<server wallet address> # receiver
AVM_MNEMONIC="<agent 25-word mnemonic>" # payer, LOCAL ONLY
SERVER_URL=http://localhost:4021
3. Start server + run agent
pnpm server # terminal 1
pnpm agent # terminal 2: discovery → 402 → auto-pay → result
Live Deployment
Production server running on Railway:
https://mcp-pay-production.up.railway.app
Try it: GET /health or browse GET /mcp/v1/tools — then pay with any x402 client.
Verify On-Chain
Every successful call settles a real TestNet transaction:
- Facilitator dashboard: https://facilitator.goplausible.xyz/dashboard
- Lora explorer: https://lora.algokit.io/testnet
Verified settlements from actual demo runs:
| Tool | Price | Transaction |
|---|---|---|
web_research |
$0.002 | PNSX3OTXTMP5V7JEDDWD2G3H7KS3R7NTB2DO4RFXNR3URYUCPPPQ |
market_data |
$0.001 | E4OFWPYIXJNGWRYMAJIA2EZOBAQXZ7HI4UOBOFRZPLLNGPAKMZKA |
contract_audit |
$0.005 | settled (see facilitator dashboard) |
market_data (production) |
$0.001 | DUITAVSQ3TU7WNHMQSNGHXBHG6MFEIBNYJY5HAYC2NNZVG5244YQ |
API
| Endpoint | Payment | Description |
|---|---|---|
GET /health |
free | liveness |
GET /mcp/v1/tools |
free | tool catalog with schemas + prices |
POST /mcp/v1/tools/:name/execute |
x402 | execute a tool, pay-per-call |
Roadmap
- Spend-policy guard: per-agent budgets enforced by the middleware
- Creator dashboard: register your own MCP tool + price in minutes
- Bazaar listing at GoPlausible so any third-party agent can discover MCP-Pay tools
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。



