Skip to main content
Pulumi logo Pulumi logo
  1. Docs
  2. Internal Developer Platform
  3. Concepts
  4. Terraform modules

Terraform modules in the Pulumi Cloud registry

    Pulumi Cloud hosts Terraform modules as a first-class registry resource alongside packages and templates. Teams migrating from HCP Terraform can publish their existing modules to Pulumi Cloud using the same tooling they already use (the go-tfe library or the hashicorp/tfe Terraform provider) by pointing those tools at tf.pulumi.com instead of app.terraform.io. Every module version you publish is also converted into a Pulumi package. The module’s variables become typed inputs and its outputs become typed outputs, with a generated SDK in TypeScript, Python, Go, C#, Java, or YAML, an API reference on the package’s page, and a record of which stacks depend on it. Conversion is additive: existing .tf consumers keep resolving the module over the Terraform protocol.

    Before you begin

    1. You need a Pulumi Cloud account on the Enterprise or Business Critical plan. Publishing is gated to those tiers; reading and listing modules is available on any plan, so you always keep access to modules you have already published.
    2. You need the Pulumi CLI installed if you plan to consume modules from a Pulumi program.
    3. You need OpenTofu or Terraform installed if you plan to consume modules from a .tf file with tofu init / terraform init.

    Authenticate

    Every surface authenticates with a Pulumi access token. It is the bearer token for everything Pulumi Cloud exposes over the HashiCorp protocol: the publish API, the state backend, and the module registry.

    • Publishing: the go-tfe client and the tfe provider take your Pulumi access token wherever they expect a TFE token today. See Publish a module.

    • Consuming from a Pulumi program: run pulumi login. pulumi package add passes the token through, so there is no separate registry login.

    • Consuming from plain OpenTofu or Terraform: set the host token. OpenTofu and Terraform derive the variable name from the host by replacing dots with underscores (and dashes with double underscores), so tf.pulumi.com becomes TF_TOKEN_tf_pulumi_com:

      export TF_TOKEN_tf_pulumi_com=$PULUMI_ACCESS_TOKEN
      

      You can also store the token in the Terraform CLI credentials file (~/.terraform.d/credentials.tfrc.json). Self-hosted installations use the same scheme with their own host.

    Publish a module

    Pulumi Cloud’s publish API is wire-compatible with HCP Terraform’s private registry. Existing HCP migration tooling works unmodified, pointed at the new host. The two most common paths:

    go-tfe

    Publish from Go, or from any CI pipeline that already drives HCP Terraform through this client, by pointing it at the Pulumi Cloud host:

    client, _ := tfe.NewClient(&tfe.Config{
        Address: "https://tf.pulumi.com",
        Token:   os.Getenv("PULUMI_ACCESS_TOKEN"),
    })
    

    The RegistryModules surface (client.RegistryModules.Create, CreateVersion, UploadTarGzip, etc.) accepts the same payloads it accepts against app.terraform.io.

    hashicorp/tfe Terraform provider

    Publish from HCL, if you manage your registry modules declaratively with OpenTofu or Terraform:

    provider "tfe" {
      hostname = "tf.pulumi.com"
      token    = var.pulumi_access_token
    }
    
    resource "tfe_registry_module" "vpc" {
      organization = "acme"
      ...
    }
    

    Module layout

    At publish time Pulumi Cloud reads the standard Terraform module structure:

    • Root .tf files (any filenames) define the module’s inputs, outputs, and required providers.
    • modules/<name>/ subdirectories are parsed the same way as the root and can be consumed as submodules.
    • examples/<name>/ subdirectories and the README.md are captured at publish.

    Migrating from HCP Terraform

    If you publish from CI today, the move is a host change. Point your existing pipelines at tf.pulumi.com, supply a Pulumi access token, and your publish steps run unchanged. Reading and listing modules work on any plan, so you keep access to modules you have already published regardless of your plan.

    Module names

    Pulumi Cloud uses the same <namespace>/<name>/<system> address form as HCP Terraform. The namespace is your Pulumi organization, and the system is the segment HCP Terraform calls the provider: what the module provisions, such as aws, azurerm, or kubernetes. One rule is stricter: the module name must match [a-z0-9][a-z0-9-]*, so it starts with a letter or digit and underscores are rejected at publish. A module that HCP hosts under a name like control_tower_account_factory has to be renamed to control-tower-account-factory before you publish it. Uppercase in the name is lowercased automatically.

    What happens when you publish

    Publishing a module version also converts it into a Pulumi package, with no extra step on your part. The package is named after the module: <name>-<system>, published under the same namespace and registry source, so a module published as acme-corp/vpc/aws produces a package called vpc-aws.

    Conversion runs per version, so a module can have some versions with packages and some without. The package’s page in Pulumi Cloud shows which versions have converted and gives you the command to install one.

    Consume from a Pulumi program

    Once a version has converted, install it by package name:

    pulumi package add <name>-<system> [<version>]
    

    The module is a multi-language component: its variable blocks become typed inputs, its output blocks become typed outputs, and Pulumi generates an SDK in the language your project uses. The version you pass is persisted in Pulumi.yaml, so pulumi install regenerates the same pinned version.

    The resources the module creates appear individually in previews and in the resource graph rather than as one opaque unit. In Pulumi Cloud the package gets the same treatment as any other: an API reference generated from the module’s variables and outputs, and usage tracking recording which stacks depend on it and which of those are behind the latest version.

    Usage tracking only counts consumption through the converted package. A stack or workspace that consumes the module over the Terraform protocol does not report a dependency, so it does not appear in the usage columns or on the package’s “Used by” tab.

    Installing a converted package requires Pulumi CLI 3.248.0 or newer. See Download & Install Pulumi to install or upgrade.

    If a version has no package

    Installing by package name is the path to reach for. If a version you need has no package, you can convert the module locally instead, against the module address rather than the package name:

    pulumi package add hcl module tf.pulumi.com/<namespace>/<name>/<system> [<version>]
    

    hcl is a parameterized provider. The module keyword selects module mode, followed by the module address and an optional version. Omit the version to resolve the latest published version; pass one to pin it.

    This converts the module on your machine, using your local hcl provider, and generates an SDK for your project. Nothing is published: there is no package in the registry, and so no package page, no API reference, and no usage tracking, and a teammate who needs the module runs the same command rather than installing what you produced. A module the registry could not convert may well fail here for the same reason.

    Both commands resolve using your Pulumi credentials. See Use a Terraform Module in Pulumi for examples.

    Consume from OpenTofu or Terraform

    Reference the module in a .tf file:

    module "vpc" {
      source  = "tf.pulumi.com/<namespace>/<name>/<system>"
      version = "1.2.3"
    }
    

    tofu init and terraform init resolve and download the module from Pulumi Cloud using the token you set above.

    Submodules are referenced with the standard //modules/<name> source syntax:

    module "private_subnet" {
      source  = "tf.pulumi.com/<namespace>/<name>/<system>//modules/<submodule>"
      version = "1.2.3"
    }
    

      The infrastructure as code platform for any cloud.