Safety

The Vultr Model Context Protocol (MCP) Server is read-only. It can view your account's resources but cannot create, modify, or delete anything, and it has no tools that spend money. If you ask an AI assistant to resize a server through the MCP server, the assistant reports that it cannot.

How the Server Protects Your Account

• Preview: The Vultr MCP Server is available in public preview. The tool surface and endpoint structure may change at general availability, so forward compatibility is not guaranteed.
• Read-only: Only read operations are exposed. Operations that change state even though they use a read method, such as purging a CDN cache, are removed as well.
• Your credential is the boundary: Every call runs as you, and Vultr enforces what your credential may do. The server cannot see anything that you cannot see.
• No identity management: Tools for API keys, users, Identity and Access Management (IAM), System for Cross-domain Identity Management (SCIM), organizations, and OpenID Connect (OIDC) are left out, so an assistant cannot read or change who has access to your account.
• Arguments are checked: A tool called with an argument that it does not have is refused, with a list of the arguments that it takes. A mistyped filter cannot quietly return everything.
• Errors come from Vultr: When api.vultr.com returns an error, you see its message. A 5xx error means that the Vultr API had a problem, so try again later. A 401 or 403 error means that Vultr refused the request.

Control Access

Connected AI assistants can see account and resource details, so enable MCP only for AI assistants and entities that you trust. An organization administrator decides whether MCP is available and who can use it. To turn MCP on or off for the organization, see Admin Controls. To remove access for a single user, detach the Use MCP permission policy from that user, group, or role.