Show HN: Spacelift Intent MCP – Cloud Infra Provisioning for Your AI Agent (github.com)

Spacelift Intent is an open-source MCP server that lets infrastructure engineers (or LLM-driven agents) describe desired cloud infrastructure in natural language and provisions resources by calling provider APIs directly — bypassing the traditional Terraform/OpenTofu configuration layer. It’s experimental but targets a key workflow shift: enable MCP hosts (Claude Desktop, VSCode/Claude Code) and LLMs to drive deterministic provider plugin calls, so agents can create, update, import, refresh and delete resources without producing intermediate IaC files. There’s also a managed version in the Spacelift Platform. Technically, Intent runs as an stdio MCP server (Docker, Homebrew or binary), uses OpenTofu providers, and requires provider credentials via environment variables. It persists all control-plane state in a local SQLite database (state_records, operations, dependency_edges, timeline_events, etc.), exposes provider-discovery and resource lifecycle tools (provider-search, lifecycle-resources-create/update/delete/import/refresh), and logs every tool call as an operation. Important implications: it simplifies agent-driven provisioning and rapid iteration, but centralizes state in a local DB (must mount a persistent volume or back up DB) and exposes risky operations (delete/eject/dependencies) that need strong access controls and audit trails. Useful for prototyping LLM-first infra automation, but treat as experimental for production until state management, RBAC and safety guards mature.