mcp-x402-accept
Proxies an existing MCP server to add an x402 paywall, charging AI agents per tool call without modifying the upstream server.
README
@cryptoapis-io/mcp-x402-accept
Put an x402 paywall in front of an MCP server you already run — charge AI agents per tool call without changing a line of that server's code.
agent → mcp-x402-accept (your process) → your existing MCP server
The proxy is an MCP server to the agent and an MCP client to yours. It mirrors every tool your server exposes; the ones you price are paywalled, the rest pass through free.
When to use this (and when not to)
| Your situation | Use |
|---|---|
| You can edit your MCP server (Node/TS) | x402-merchant-sdk/mcp — three lines around your handler, nothing in the data path |
| You cannot — third-party, closed-source, or not Node | this proxy |
If you can edit your server, do that instead. It is simpler and strictly fewer moving parts. This package exists for the case where that is not an option.
Why self-hosted only
You run this process. We never do, and we do not offer a hosted version.
That is a deliberate product decision, not a default you can flip. A proxy sits between an agent and your server, so a hosted one would mean CryptoAPIs holding your upstream server's credentials, sitting in the cleartext path of your tool arguments and results, and owing you uptime on your own revenue. For a merchant whose product is the data flowing through those calls, that is not a reasonable thing to ask. Self-hosting removes the question rather than answering it.
What that guarantees, structurally:
- Tool arguments and results are opaque to the proxy. It reads the tool name (is this priced?)
and
_meta["x402/payment"](is it paid for?) — nothing else. They are never parsed, logged, buffered, or cached. - Your upstream credentials are your own env vars, in your own process.
- The only outbound calls are to the CryptoAPIs facilitator, to
/verifyand/settlethe payment itself.
Install
npm install -g @cryptoapis-io/mcp-x402-accept
Configure
{
"upstream": {
"transport": "stdio",
"command": "node",
"args": ["/path/to/your-mcp-server.js"],
"env": { "YOUR_UPSTREAM_TOKEN": "${YOUR_UPSTREAM_TOKEN}" }
},
"apiKey": "${CRYPTOAPIS_API_KEY}",
"payTo": "0xYourReceivingAddress",
// Only these are charged for. Everything else your server exposes stays FREE.
"pricedTools": {
"premium_data": {
"network": "eip155:8453",
"asset": "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913",
"amount": "10000"
}
}
}
${VAR} reads from the environment, so the config file itself holds no secrets and is safe to commit.
amount is in atomic units — USDC has 6 decimals, so "10000" is $0.01. Getting this wrong by
10^6 is the easiest mistake here.
An HTTP upstream instead of stdio:
"upstream": {
"transport": "http",
"url": "https://your-server.example.com/mcp",
"headers": { "Authorization": "Bearer ${YOUR_UPSTREAM_TOKEN}" }
}
To accept several assets or networks for one tool, pass an array — the agent picks one. Offer only
what GET https://ai.cryptoapis.io/x402/merchant/supported lists (public, no key); anything else is
advertised and then rejected at /verify.
Run
cryptoapis-mcp-x402-accept --config ./x402-accept.config.json
Then point agents at the proxy instead of your server. In Claude Code:
claude mcp add my-paid-tools -- cryptoapis-mcp-x402-accept --config /abs/path/to/config.json
Everything is validated at boot — the config, the upstream connection, and whether the tools you priced actually exist upstream. A typo'd tool name is a hard failure, not a silent revenue hole: the alternative is believing a tool is paywalled while it quietly serves for free.
What an agent sees
| Step | Result |
|---|---|
| Unpaid call to a priced tool | isError: true carrying PaymentRequired in both structuredContent and content[0].text |
| Agent pays | Signs locally, retries with the payload in _meta["x402/payment"] — raw JSON, no base64 |
| Paid call | Your tool's real result, with the receipt in _meta["x402/payment-response"] |
| Call to an unpriced tool | Forwarded, free, unchanged |
Your server is never called until payment has settled. The proxy charges and then forwards — it does not forward and then bill. An unpaid call cannot reach your server at all, so it cannot be made to do the work for free.
Priced tools advertise [paid: x402] in their description so an agent knows the cost before
calling. An agent that discovers a price only by being refused has already wasted a round-trip.
Paying it
Agents pay with @cryptoapis-io/mcp-x402-pay
or, in code, @cryptoapis-io/x402-buyer-sdk/mcp.
Notes
- Requires a CryptoAPIs API key with the
X402_FACILITATORfeature. "settle": falseverifies without settling — useful while wiring up, but advisory only: the buyer proved they could pay, not that they did. Never ship it.- Test end-to-end on Base Sepolia (
eip155:84532) before charging real money. - Implements the x402 v2 MCP transport (spec), sharing one implementation with the merchant SDK — a spec fix lands in both.
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。