Claude Code Plugin

The threatcl/claude-plugin marketplace brings Threatcl into Claude Code. It publishes two plugins:

• threatcl-cloud: a skill, five workflow slash commands, and the Threatcl Cloud MCP server.
• threatcl-lsp: an opt-in language server plugin, giving Claude live diagnostics while it edits HCL threat models.

Install

terminal

$ claude plugin marketplace add threatcl/claude-plugin
$ claude plugin install threatcl-cloud@threatcl

You can also install from inside Claude Code with the /plugin command.

The first time you use a Threatcl Cloud feature, Claude Code opens a browser to complete OAuth. Pick your organization and you’re done, with no tokens to copy around.

Prerequisites

• Claude Code. These plugins are Claude Code specific. For Codex and other agents, follow the agent setup guide instead.
• The threatcl CLI, for write operations and local file work. See Installing threatcl.
• A Threatcl Cloud account, at threatcl.com.

Slash commands

Command	Argument	Purpose
/threat-review	<model-slug-or-id>	A structured security review of a cloud model: unmitigated threats, STRIDE coverage, policy evaluation, library-fit suggestions and prioritized next actions
/threat-for-code	<path-or-diff-range>	Analyze code and suggest threats from your org library, plus novel threats worth modeling, with HCL snippets
/threat-drift	[diff-range]	Check the model’s claims against current code. Defaults to main...HEAD
/threat-ci	<github-actions|gitlab-ci|pre-commit>	Scaffold CI integration, with cloud validate on PR and cloud policy evaluate on merge
/threat-hcl-new	<model-name>	Scaffold a new HCL threat model with the backend block pre-populated

/threat-for-code is the discovery command, asking what threats apply to this code. /threat-drift is the diagnostic one, checking whether what the model already claims is still true. For drift on every pull request rather than on demand, see the drift action.

The skill

The bundled threatcl skill auto-triggers when Claude detects threat-modeling intent, and steers it toward the right tool for the job: the MCP server for reads, the CLI for writes and local files.

MCP server

The plugin declares the hosted Threatcl Cloud MCP server, which gives Claude read access to your organization:

• list_threat_models and get_threat_model, for models and their HCL
• search, across threat models and the threat, control and information asset libraries
• list_library_items and get_library_item, for library lookups with version detail
• get_usage_analytics, for library usage statistics
• update_threat_model_status, to move a model between statuses

Authentication is handled by Claude Code over OAuth. For everything else, push, validate, library import and policy edits, the skill falls back to the threatcl CLI, which uses your stored token.

Two different MCP servers

This is not the same thing as the threatcl mcp command. That one runs locally over stdio and exposes the HCL files in a directory you point it at, with no account required. This one is hosted, reaches your Threatcl Cloud organization, and authenticates with OAuth. Local files, or cloud data: pick the one that matches the work.

Endpoint

The server’s endpoint is declared in the plugin’s .mcp.json. If your organization is on a different deployment, edit that file after install. The CLI’s endpoint is separate, and is saved with each token or overridden with THREATCL_API_URL. See Authentication.

The threatcl-lsp plugin

threatcl-lsp is a separate, opt-in plugin that wires the threatcl lsp language server into Claude Code. Claude then gets diagnostics pushed into its context each time it edits an HCL threat model: syntax errors, unknown blocks and attributes, missing required attributes and invalid enum values. It can catch and fix invalid HCL in the same turn it writes it.

terminal

$ claude plugin install threatcl-lsp@threatcl

It requires threatcl 0.5.0 or newer on your PATH. The plugin only tells Claude Code how to launch the server, it doesn’t bundle it.

The *.hcl caveat

Claude Code matches language servers on a file’s final extension segment only. A file named model.tm.hcl resolves to .hcl, so this plugin necessarily claims all *.hcl files. Two consequences:

• It also attaches to Terraform, Packer and Nomad HCL, where it may report spurious “unknown block” diagnostics.
• Only one server can own .hcl. If you also install a Terraform or HCL language server plugin, whichever Claude Code loads first wins, and the other is dropped.

That’s why it isn’t bundled into threatcl-cloud. Install it if you mostly work in threatcl HCL, and skip it if your repos are mostly Terraform.

Note that this differs from the advice for editors with finer-grained matching, where naming your files *.tm.hcl and scoping the server to that suffix avoids the collision entirely. See LSP.