← Reddit

Built an MCP that lets Claude run an entire EU cloud (Hetzner), with a cost guard

Reddit · mjiqbal · June 8, 2026
A developer created hetzner-mcp, a Model Context Protocol tool that enables Claude to provision and manage resources on Hetzner's EU cloud platform, addressing a gap in existing MCP offerings. The tool comprises 32 typed tools across Cloud, Storage Boxes, and dedicated servers with safety mechanisms that display pricing before resource creation and require confirmation for deletions. All 39 API endpoints were validated through testing, including a full stack deployment and teardown completed in seconds for under a cent.

Detailed Analysis

A developer has released hetzner-mcp, an open-source Model Context Protocol (MCP) integration that enables Claude to programmatically manage infrastructure on Hetzner, a popular European cloud provider known for competitive pricing and EU-based data residency. The tool exposes 32 distinct tools spanning Hetzner's Cloud platform, Storage Boxes, and Robot dedicated server management, with three generic endpoint-covering tools complemented by typed shortcuts for common operations. The developer validated all 39 endpoints live before release and demonstrated the tool's practical capability by having Claude provision and tear down a full stack for under one cent.

The project addresses a meaningful gap in Claude's existing MCP ecosystem. While integrations for major cloud providers like AWS and platforms like Vercel already existed, Hetzner — widely used among European developers and privacy-conscious organizations — had no corresponding MCP support. This left users who rely on Hetzner for GDPR compliance or cost optimization without AI-assisted infrastructure management, even as counterparts on other platforms could delegate provisioning tasks directly to Claude.

A notable engineering decision in the project is its cost guard mechanism. The MCP is designed so that Claude cannot create any billable resource without first surfacing the live price and requiring explicit user confirmation. Destructive operations like deletions similarly require confirmation, while read operations are unrestricted. This design reflects a broader challenge in agentic AI tooling: enabling autonomous action while maintaining meaningful human oversight over operations with real-world financial or data-integrity consequences. The approach operationalizes a principle of asymmetric caution — free actions flow freely, costly or irreversible ones pause for human approval.

The release fits into a broader trend of community-driven expansion of Claude's MCP ecosystem, where developers are independently building connectors to services that Anthropic has not yet formally integrated. MCP, as an open protocol, enables this kind of decentralized extension, and infrastructure management represents one of the highest-stakes domains for such tooling. The community response to projects like hetzner-mcp indicates growing appetite for Claude to function as a genuine infrastructure operations agent, not merely a coding assistant, particularly among European developer communities with specific regulatory and cost constraints that make Hetzner a preferred environment over US-headquartered hyperscalers.

Read original article →