Home / AI Agent Hosting
AI agent hosting · MCP-native · No KYC

Hosting your AI agent buys itself.

A KYC-free offshore VPS, RDP or GPU box your agent provisions over MCP or REST, pays for with crypto, and tears down — no card, no human in the loop, live in 60 seconds.

MCP
REST
No KYC
Crypto only
Dry-run
x402
Why agentic hosting

Most clouds were built for people. This one answers to code.

An autonomous agent stalls the moment a flow demands a credit card, a phone number or an ID upload. ServPrivacy has none of those gates — so an agent can go from nothing to a running server without a human ever clearing a check.

No human in the loop

Email + password are not required — a 16-character token is the whole identity floor. There is no KYC step that needs a person to clear it, so a headless agent runs the full flow alone.

Payment a machine can make

Cards assume a person with a billing address. Agents settle in crypto: top up a prepaid balance once, then every deploy draws from it. The cost is explicit and bounded — nothing kept on file to bill again.

Built to be discovered

The surface is machine-first: a hosted MCP server, an OpenAPI 3.1 spec and an A2A agent card under /.well-known/. Point a client at it and the tools register themselves.

What your agent can do

Every order, as a tool it can call

Browse & quote

Enumerate VPS, dedicated, RDP and GPU plans across 7 jurisdictions and price any configuration — all public, no token to look.

Top up in crypto

Open a crypto invoice for any of 20 coins, poll it to confirmation, and collect the automatic bonus credit.

Deploy & control

Order a VPS, RDP or GPU box from balance, read back credentials, then reboot, reinstall or shut it down — all via the API.

Configure GPU at deploy

Pre-install AI stacks (vLLM, ComfyUI, Ollama) and pre-download models (Llama-3.3-70B, FLUX) in the order itself.

The full surface — tools, schemas, OAuth, x402 and the public agent directory — lives on the agent docs page, and powers a private GPU box for AI workloads.

How it works

From a prompt to a running box. Three steps.

01

Connect the MCP server

Point Claude, Cursor or any MCP client at servprivacy-mcp (or the hosted endpoint). The tools register automatically.

02

Let the agent pay

It opens a crypto top-up and waits for confirmation. You set the budget; the prepaid balance is the ceiling it cannot exceed.

03

It provisions the server

The agent deploys the box you described, reads back the credentials once, and hands you a running machine — or tears it down when done.

Open-source MCP server

Install once, then ask in plain language

The servprivacy-mcp server (MIT, on PyPI and GitHub) maps the whole API to MCP tools, so your assistant can compare plans, top up in Monero or Bitcoin and deploy a server — including a GPU box with models pre-loaded — conversationally.

uvx servprivacy-mcp
# or
pip install servprivacy-mcp
Agentic vs identity-bound

Why agents stall on a normal cloud. And don't here.

DimensionServPrivacyIdentity-bound cloud
Signup✅ Token, no human review❌ Account + verified card, often ID
KYC✅ None, ever❌ Identity + billing address
Payment✅ Prepaid crypto balance❌ Card on file, monthly invoice
Provisioning✅ MCP + 50+ REST endpoints⚠️ REST behind a human account
Rehearsal✅ Built-in dry-run mode⚠️ Trial credit, if any
Teardown✅ Agent cancels; nothing recurring❌ Subscription tied to a card
Jurisdiction✅ 7 offshore jurisdictions⚠️ Provider home jurisdiction
FAQ

Agents and servers, answered

01 Can an AI agent really buy a server on its own?

Yes — end to end. ServPrivacy ships a public Agent API (REST + a hosted MCP server) so an agent can create an account, top up a balance with cryptocurrency, and deploy a VPS, Windows RDP or GPU box with no human review. The identity floor is the same for a person and a model: a 16-character token, no KYC. A server is active in under 60 seconds.

02 How does the agent pay without a card?

The model is balance-only. The agent tops up a USD balance once with crypto (Bitcoin, Monero or 18 other coins, from $30, with up to +100% bonus credit), then every deploy debits that prepaid balance. There is no card on file and nothing recurring — if the balance is short, the order returns HTTP 402 with the exact amount to top up, which the agent can act on automatically.

03 Which clients and frameworks are supported?

Claude Desktop, Claude Code, Cursor, Cline, Continue and any MCP-aware client, plus plain REST for everything else. There is an official open-source MCP server — install it with "pip install servprivacy-mcp" or "uvx servprivacy-mcp" — that exposes the whole catalogue, billing and server lifecycle as tools your assistant can call in natural language.

04 How do I stop an agent from over-spending?

Spending is bounded by design. The balance is prepaid, so there is no card to run up. GPU orders can carry an auto-shutdown timer (6h–7d) so a forgotten box self-destructs. And the whole flow has a dry-run mode (?dry_run=1) so an agent — or you — can rehearse a top-up and deploy without spending a cent or provisioning anything.

05 Is it really no-KYC for agents too?

Yes. No email, no name, no phone, no ID — for humans or agents alike. The account token is issued automatically on the first crypto top-up and shown once. No traffic, connection or DNS logs are kept at the hypervisor level. The warrant canary documents exactly what is and is not collected.

06 What can the agent actually do through the API?

Browse the catalogue (VPS, dedicated, Windows RDP, GPU), quote a configuration, open and poll a crypto top-up, deploy and act on servers (reboot/shutdown/reinstall), read credentials, and check anonymous domain availability. GPU deploys accept deploy-time AI config: pre-installed stacks (vLLM, ComfyUI, Ollama) and pre-downloaded models (Llama-3.3-70B, FLUX, DeepSeek-R1). Every action has a REST endpoint and an MCP tool.

Give your agent the keys

Wire it to the MCP server, fund a small crypto balance, and let it stand up a server the moment it needs one.

Read the agent docs