CloudPloy

CloudPloy is a deployment control plane built for AI agents: it wraps app creation, domain setup, and SSL signing into roughly 30 MCP tools, so one sentence in Claude Code or Cursor is enough to ship your code live. Your apps run in your own AWS account or on any Ubuntu/Debian server, and it never calls a large language model.

Core capabilities

• MCP native: listed in the official MCP Registry; after OAuth authorization your own agent calls it directly
• Bring your own compute: your own AWS EC2 account, or any fresh Ubuntu 22.04+ / Debian 12+ machine, with compute billed at the provider's list price and zero markup
• Framework detection: detect_app reads the repository to identify the framework and port, and Nixpacks covers Node, Python, Go, Rust, PHP and more
• Domains and certificates: verify_domain and provision_ssl chain into Let's Encrypt issuance, and you can also connect Cloudflare to create DNS records
• Permission boundaries: tiered read / deploy / provision / destroy tokens; destruction requires confirm=true; you can call estimate_cost first for a quote

Where it fits Independent developers who live inside AI editors, cost-conscious teams with their own VPS, agencies taking client work, and anyone who wants a flat subscription instead of Railway's usage-based billing. Workloads that need autoscaling or load balancing are not a fit yet.

Unique strengths Most competitors treat agents as an afterthought bolted onto an existing API; CloudPloy treats MCP as a first-class citizen: registry listing, OAuth 2.1 + PKCE, browser-based authorization, and secrets never enter the conversation. It charges only for the control-plane subscription, you pay the cloud bill yourself, and there is no markup and no resale.

Editorial take The value proposition is clear: a flat subscription starting at $9.99 replaces unpredictable usage bills, and the permission boundaries are drawn for agents up front. It suits people who bring their own machines and spend the day in Claude Code. But sign-up is still invite-only, and with no SLA and no autoscaling we would not run production-critical systems on it; for a managed platform that works out of the box, Railway is less work.