免 KYC
只需一个邮箱地址或 Telegram 用户名即可下单。我们不要求护照、自拍、电话号码或地址证明。姓名可选，且不做核验。
路线图 · 计划中
Infrastructure your AI agents can buy themselves
We're building a REST API and an MCP server so autonomous agents can quote, order, pay for and manage servers with a crypto wallet, within limits you set. It isn't live yet. This page documents the design so you can plan against it.
本页面目前仅提供英文版本。
REST API
Token-authenticated JSON endpoints to quote, order, renew, reboot and reinstall. Idempotency keys on every write.
MCP server
The same actions exposed as tools for any MCP-capable agent runtime, so a model can provision its own compute.
Wallet billing with guardrails
Agents pay from a prepaid crypto balance. You set per-token spend caps, allowed products and allowed locations.
计划中
Endpoints
Base URL will be https://api.mercurehosting.com. It isn't reachable yet.
| 方法 | 路径 | 用途 |
|---|---|---|
| GET | /v1/catalog | List products, plans, locations and list prices |
| POST | /v1/quotes | Price a plan + term + coin without committing |
| POST | /v1/orders | Create an order and receive a crypto invoice |
| GET | /v1/orders/{ref} | Poll invoice and provisioning status |
| GET | /v1/servers | List servers owned by the token |
| POST | /v1/servers/{id}/actions | reboot · power_off · power_on · reinstall · rescue |
| POST | /v1/servers/{id}/renew | Extend the term, paid from balance or new invoice |
| GET | /v1/balance | Prepaid crypto balance and spending limits |
计划中
Example: quote, then order
Request · create quote计划中
POST https://api.mercurehosting.com/v1/quotes # PLANNED — not live
Authorization: Bearer mh_live_••••••••
Content-Type: application/json
{
"plan": "IS-L",
"term": "6m",
"coin": "XMR"
}响应计划中
200 OK
{
"plan": "IS-L",
"location": "iceland",
"term": "6m",
"monthly_usd": 34.50,
"effective_monthly_usd": 22.43,
"discount": 0.35,
"total_usd": 134.58,
"coin": "XMR",
"quote_id": "q_7Hk2…",
"expires_at": "2026-11-01T12:00:00Z"
}Request · create order计划中
POST https://api.mercurehosting.com/v1/orders # PLANNED — not live
Authorization: Bearer mh_live_••••••••
Idempotency-Key: 4b1e9c1a-…
{
"quote_id": "q_7Hk2…",
"os": "debian-12",
"ssh_keys": ["ssh-ed25519 AAAA… agent@runner"],
"hostname": "inference-01",
"pay_from": "balance" // or "invoice"
}响应计划中
201 Created
{
"ref": "MH-261031-9F2C1A",
"status": "awaiting_payment", // → paid → provisioning → active
"invoice": {
"coin": "XMR",
"amount": "…", // set by the payment gateway
"address": "…",
"expires_at": "2026-11-01T12:00:00Z"
},
"server": null // populated once active
}Prices in these examples follow today's list prices and term discounts. Field names may change before release.
计划中
MCP server
One server exposing five tools. Agents discover them automatically.
Tool manifest (abridged)计划中
{
"tools": [
{ "name": "mercure_list_plans", "input": { "product": "vps|dedicated|rdp|gpu", "location?": "string" } },
{ "name": "mercure_quote", "input": { "plan": "string", "term": "1m|3m|6m|12m", "coin": "BTC|XMR|…" } },
{ "name": "mercure_create_order", "input": { "quote_id": "string", "os": "string", "ssh_keys": "string[]" } },
{ "name": "mercure_order_status", "input": { "ref": "string" } },
{ "name": "mercure_server_action", "input": { "server_id": "string", "action": "reboot|reinstall|power_off" } }
]
}Client configuration计划中
{
"mcpServers": {
"mercurehosting": {
"command": "npx",
"args": ["-y", "@mercurehosting/mcp"], // PLANNED package name
"env": { "MERCURE_API_TOKEN": "mh_live_…" }
}
}
}Guardrails we're building in
Spend caps
Per-token daily and monthly USD limits.
Scoped tokens
Read-only, order-only or full-control scopes.
Location allow-lists
Restrict an agent to the jurisdictions you approve.
Human-in-the-loop
Optionally require your approval above a set amount.
现在
What you can run before the API exists
The endpoints above are a design. Ordering today is the same form a person uses. An agent can live on the server once that server exists.
A VPS for the agent
Most tool-calling loops fit in 4–8 GB of RAM. Add memory if a vector database shares the box. Full root, so the framework is whatever you install.
A GPU if the model stays local
RTX 4090, RTX 5090 or H100 in Iceland, Romania, Switzerland or the Netherlands. Prompts to that model stay on hardware you rent. We don't log them.
A person pays the invoice
Create the order, pick a coin, pay the 60-minute invoice. There is no live call an agent can make to buy a server by itself. Wallet billing with spend caps ships with the API, not before.
适合谁
People planning a long-running agent
Framework authors
You want a stable address and a machine that is still there tomorrow, in a country you chose, without a card on the account.
Teams keeping prompts in-house
Pair the agent VPS with a GPU plan. The uncensored AI page explains the limit: we don't filter the model, and the AUP still applies.
Builders reading the spec early
The shapes on this page are here so a client can be sketched. Don't point production traffic at api.mercurehosting.com. It is not in service.
相关
Order the machine while the API is still a design
AI agent hosting
The product page: VPS now, GPU optional, API described as planned.
Uncensored AI hosting
Open-weight models on your GPU, without a third-party filter.
No-KYC GPU servers
The cards, the countries, and the monthly price.
VPS catalog
Where the agent process actually runs today.
Order a VPS
The live checkout. Access token, prepaid credit, then a queued machine.
FAQ
Including the straight answer that this API is not live.
常见问题
Questions about agents and the API
Can I call the API today?
No. Every endpoint and tool on this page is marked planned. The base URL is not in service.
How do I host an agent right now?
Create an access token, add credit, and order a VPS from the catalog. Install your framework on the machine. Add a GPU plan if you want the model on your own hardware.
Will agents be able to spend without a limit?
The design includes per-token spend caps, location allow-lists and an optional human approval. None of that is enforceable until the API exists, because the API is the thing that would issue the token.
Who is responsible for what the agent does?
The customer who ordered the server. The AUP applies to software the same way it applies to people.
Do you log prompts on a GPU I rent?
No. We don't inspect the workload. A third-party model API you choose to call has its own logging, which is outside the server.
可接受使用政策适用于所有服务
隐私不是滥用的许可。CSAM、恶意软件、网络钓鱼、垃圾信息、僵尸网络、欺诈以及在服务器所在地违法的任何内容均被禁止，违反这些规则的服务将被暂停。我们遵守服务器所在司法管辖区发出的有效法院命令。请阅读完整的 服务条款与可接受使用政策.
Agents are bound by the same Acceptable Use Policy as people, and the customer who ordered the server is responsible for what their agents run.
Building an agent that needs compute?
Tell us what you're building and we'll add you to the early-access list.
每个页面都有
离岸 VPS 一目了然
离岸 VPS 是一台位于你所选国家的 KVM 服务器，用邮箱或 Telegram 用户名下单，以加密货币付款。国家决定了机器适用的法律。来自其他地方的投诉不会导致自动下架，可接受使用政策在七个机房均适用。
加密货币结算
使用 BTC, ETH, USDT, USDC, LTC, DOGE, TRX, SOL, BNB, XRP, BCH 付款。价格以美元标示。先在钱包中充值，再用余额下单。无银行卡，无银行账户。转账到该地址，然后等待工作人员确认。
七个司法管辖区
冰岛、巴拿马、摩尔多瓦、罗马尼亚、瑞士、荷兰、俄罗斯。你在下单时选择国家，任何投诉都依据该国法律审理。在你支付长期费用之前，各机房页面会讲清利弊。
有明确清单的隐私
我们不记录 VPS 收发的流量，也不保存磁盘内容。存在的计费记录包括：一种联系方式、订单、付款凭证、IP 和电源状态。隐私政策列出了完整清单。
坦诚处理通知
美国 DMCA 通知不会导致自动下架。投诉依据 VPS 所在国家的法律审核。该国的有效法院命令会被执行。依据该国法律构成的版权侵权仍然违反可接受使用政策。
KVM、root 与 VNC
每台 VPS 都是 KVM，拥有完整 root，网络中断时还有 VNC 控制台。可安装模板或上传自己的 ISO。这不是被面板挡在中间的容器。
VPS 本身还包括：约 3,500 MB/s 级别的 NVMe、合理使用的不限流量端口、季付 −25%、半年付 −35%、年付 −50%，以及与典型 KYC 主机商的对比。我们不公布实测可用率，周期折扣之外也没有优惠码。