Servprivacy Mcp

ServPrivacy MCP server

> Deploy KYC-free, crypto-paid offshore servers — VPS, dedicated, Windows RDP and GPU AI boxes — straight from an AI agent. Claude, Claude Code, Cursor, Cline, Continue, or your own MCP client.

[](LICENSE) [](https://www.python.org/) [](https://modelcontextprotocol.io) [](https://github.com/astral-sh/ruff)

servprivacy-mcp exposes the ServPrivacy Agent API as Model Context Protocol tools, so an AI assistant can run the whole hosting workflow in natural language: compare plans, pick an offshore jurisdiction, top up a balance with Monero or Bitcoin, and spin up a server — or rent an H100 with vLLM and Llama-3.3-70B pre-loaded — without ever touching a dashboard or showing an ID.

ServPrivacy is a privacy-first offshore hosting provider operating in 7 jurisdictions (Iceland, Panama, Moldova, Romania, Switzerland, Netherlands, Russia). Accounts need no email, no phone, no name and no KYC — a 16-character token issued on your first crypto payment is the entire identity floor — and billing is crypto-native across 20 coins. The API was built for agents from day one (a public OpenAPI 3.1 spec, an agent capability card and a hosted MCP endpoint), which makes it a natural fit for a local stdio MCP server too.

Why this exists

Agentic workflows increasingly need infrastructure on demand, and most clouds stop an autonomous agent dead at a credit-card form or an ID upload. ServPrivacy has neither. An agent can stand up a throwaway box for a scraper, a Tor relay, a self-hosted VPN, a crypto full node or a seedbox — paying with Monero or Bitcoin against a host that demands no identity documents and ignores DMCA in DMCA-resistant jurisdictions.

The GPU AI angle

The standout for agent builders is GPU AI hosting. Order an RTX 4090, RTX 5090, H100 or dual-H100 box and configure it at deploy time: pre-install AI stacks (vllm, comfyui, ollama-webui, axolotl, unsloth, …), pre-download HuggingFace models (llama-3.3-70b, deepseek-r1, flux-dev, sdxl-base, …), attach an hf_token for gated weights, expose a Let's Encrypt HTTPS endpoint, and set an auto_shutdown_hours timer so a forgotten experiment can't drain the balance. CUDA + cuDNN are preinstalled. Iceland runs on 100% renewable geothermal/hydro power — the lowest-carbon offshore AI compute on the market.

Install & run

The server speaks MCP over stdio, so any MCP client launches it as a subprocess. The simplest path uses uv (no manual virtualenv):

uvx servprivacy-mcp

Or install it into your environment with pip / pipx:

pip install servprivacy-mcp
# the console script is then on your PATH:
servprivacy-mcp

> Want the latest unreleased code? Install straight from source instead: > uvx --from git+https://github.com/ServPrivacy/servprivacy-mcp servprivacy-mcp

Configuration

| Variable | Required | Default | Purpose | | --- | --- | --- | --- | | SERVPRIVACY_TOKEN | for authenticated tools | — | Your account token (AAAA-BBBB-CCCC-DDDD), auto-issued on your first create_topup and shown once — store it. | | SERVPRIVACY_BASE_URL | no | https://servprivacy.com | Override the API base URL (testing/self-host). |

Discovery tools — agent_help, list_catalog, list_locations, quote, topup_bonus, check_domains — and create_topup / create_account work with no token at all (the token is minted for you on the first top-up).

Claude Desktop

Add to claude_desktop_config.json (see [examples/](examples/)):

{
  "mcpServers": {
    "servprivacy": {
      "command": "uvx",
      "args": ["servprivacy-mcp"],
      "env": { "SERVPRIVACY_TOKEN": "AAAA-BBBB-CCCC-DDDD" }
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "servprivacy": {
      "command": "uvx",
      "args": ["servprivacy-mcp"],
      "env": { "SERVPRIVACY_TOKEN": "AAAA-BBBB-CCCC-DDDD" }
    }
  }
}

Try it

Once connected, ask your assistant things like:

• "Compare ServPrivacy's VPS tiers and tell me the cheapest one with 16 GB RAM in Iceland."
• "Rent an H100 in Iceland with vLLM and Llama-3.3-70B pre-installed, expose an HTTPS endpoint, and auto-shutdown after 24h."
• "Top up $250 of Monero (grab the bonus tier) and wait for it to confirm."
• "Deploy a Windows RDP box in the Netherlands and read me the Administrator password."
• "Is mybrand.com available, and how much vs Namecheap?" (the .com loss leader is $4.99/yr).

There's also a built-in onboard_to_servprivacy prompt that drives the full top-up → deploy flow, and a dry_run flag on create_topup/order_server to validate everything without spending a cent.

The payment model (balance-only)

ServPrivacy is balance-only — there is no direct crypto-to-server purchase:

1. create_topup — fund USD credit with crypto (min $30, max $2000). Bonus is applied automatically on confirmation (e.g. $500 → $700 credited, +40%; $2000 → $4000, +100%). See topup_bonus.
2. get_topup — poll until status: "confirmed".
3. order_server — deploy from the credit; the server is active immediately. If the balance is short it returns HTTP 402 with a topup_required block whose suggested_usd already covers the deficit and the $30 minimum — feed it back into create_topup and retry.

Tools

13 tools, mapped onto the REST API. Tools marked public need no token.

| Group | Tools | | --- | --- | | Discovery (public) | agent_help, list_catalog, list_locations, quote, topup_bonus, check_domains | | Account | create_account (public), get_account | | Billing | create_topup (public), get_topup | | Servers | order_server (public; auto-creates account), get_server, server_action |

order_server carries the GPU AI deploy-time fields — ai_stacks, pretrained_models, hf_token, ssh_public_key, public_endpoint, auto_shutdown_hours — alongside the standard type / plan / location / os / billing.

Resources

• servprivacy://catalog — full VPS/dedicated/RDP/GPU price-spec matrix (JSON)
• servprivacy://locations — the 7 jurisdictions and their legal posture (JSON)
• servprivacy://bonus — top-up bonus tiers (JSON)
• servprivacy://doctrine — the operating principles (text)

How it works

AI agent ──MCP/stdio──▶ servprivacy-mcp ──HTTPS──▶ https://servprivacy.com/api/v1

The registry in [tools.py](src/servprivacymcp/tools.py) mirrors ServPrivacy's own Agent API and binds each tool to its REST endpoint, so the surface stays in lock-step with the upstream OpenAPI spec. Authentication is a single bearer token; nothing is cached to disk.

git clone https://github.com/ServPrivacy/servprivacy-mcp
cd servprivacy-mcp
uv venv && uv pip install -e . ruff
python -m servprivacy_mcp   # starts the stdio server
ruff check .