Vultr MCP

Vultr's MCP server is generated from the platform's OpenAPI specification, so the tool surface tracks the public API rather than a hand picked subset. That gives an agent read and write coverage across compute instances, the managed Kubernetes engine, block storage and snapshots, SSH keys, load balancers, managed databases, container registry, private networks, object storage, CDN, DNS records, firewall rules, billing and account data. In practice an assistant can provision an instance, attach storage, open a firewall rule, point a DNS record at it and then read the invoice line it produced. One design choice deserves attention: identity and credential management categories, including API keys, users, IAM, SCIM, organizations and OIDC endpoints, are excluded from the generated tools by default. An agent therefore cannot mint credentials or change who has access, which removes the worst case outcome of a confused or manipulated model. Deployment works either way. Locally the server runs over stdio with a Vultr API key in the environment. Remotely it is available as a hosted HTTP endpoint or self hosted in Docker or Kubernetes, and authentication accepts both a bearer API key header and an OAuth 2.1 and OIDC browser sign in, with a verifier that handles JWTs and raw API keys side by side. Compatible clients include Claude, Claude Desktop, Cursor, VS Code and Codex CLI. It targets developers and infrastructure teams who run on Vultr and want provisioning driven from the assistant they already work in.