Pulumi MCP

Pulumi's MCP server connects an AI agent to infrastructure as code in two directions: the Pulumi Registry, which documents every resource type, input and output across the supported cloud and SaaS providers, and Pulumi Cloud, which holds the state of the stacks a team has actually deployed. The registry side answers authoring questions, letting an agent look up the exact schema for a resource instead of guessing property names. The Cloud side answers operational questions about what exists, in which stack, with which configuration. Beyond lookups it can drive the Pulumi CLI: run a preview to see the diff a change would produce, run an update to apply it, refresh state against the provider, and read stack outputs. That turns the usual write, preview, fix cycle into something an agent can iterate on with real feedback rather than compiling blind. A hosted remote endpoint and a locally run server both exist, so teams can choose whether cloud credentials leave their own machines. It is built for platform and DevOps engineers writing Pulumi programs in TypeScript, Python, Go or the other supported languages, and for anyone auditing what a stack contains before touching it. Pairing registry schemas with live stack state is the part that makes agent-written infrastructure code accurate rather than merely plausible.