What is Droost?

A collection of tools for AI-assisted Drupal development. The module reads and changes the live site; Workflow indexes project files and coordinates changes through plan, code, test and complete. What Droost is.

Is it a module or a tool?

Both, in two pieces:

  • The module (drupal/droost) runs inside your site. It lets the agent ask the site about itself and change it through gated tools.
  • The command (droost, from droost/workflow) runs beside your site, in tools/droost/. It is the brain (what the agent knows about the code) and the workflow (how each change is built). It is not a module and needs no running Drupal.

Each works without the other.

Which part do I need?

  • Building a new site? Start from the Droost CMS recipe. It brings the module and the theme.
  • An existing site? Install both: Getting started.
  • Only want the agent held to a process? The command alone.
  • Only want the agent, or an editor's AI client, to see the site? The module alone.

Which AI agents does it work with?

Any agent that can run commands. droost init writes the workflow into AGENTS.md, which Codex, Gemini and others read, and the checks run whatever wrote the code. Claude Code also gets the guard, which stops the agent from skipping a step or changing the rules part way, and a plugin to watch each run.

Can I run it on a production site?

Part of it:

  • Yes: the module's content tools, and droost_mcp for editors. Over the web each call is the account's own, held to Drupal's access.
  • No, development only: the build and developer tools (droost_build, droost_dev), Droost UI Kit, and the workflow.

The module has the table.

Does it need MCP?

No. The module's tools are Tool API tools, reached by drush droost:tool, drush tool:run, Drupal AI, or MCP. MCP is one way in, mostly for editors working from Claude Desktop or claude.ai: Over MCP.

Does the agent need Drupal running?

The command does not: it reads your project's files, answers in milliseconds, and still works when a change has broken the site. The module answers from the running site, so it needs Drupal working.

Can the agent change things on its own?

Only what you let it:

  • Module writes check category gates. Ordinary CLI writes are enabled by default unless disabled; PHP evaluation and raw SQL writes are off. Web writes require an explicitly enabled gate as well as the account's permissions. Write gates.
  • The commands that loosen the rules (changing the level, waiving a check, writing a baseline) are yours too. The guard refuses them from the agent, which gives you the exact command to run instead.

What do the levels mean?

How much each change must prove, from low (basic static checks and browser verification) to max (everything, at its strictest). high is a good default. Settings.

Does Droost send my code anywhere?

No. The module and the command run on your machine or your server, and Droost has no service of its own to send anything to. Your agent talks to its own AI provider as it always does. Two optional pieces reach out, only where you point them: the command's semantic code search calls an embeddings endpoint you name in your settings, and droost_ai calls the AI providers your site already has configured in Drupal AI.

What does it cost?

Nothing. Every Droost project is free software under the GPL-2.0-or-later.

I am on Droost 1.x or a 2.0 alpha. How do I upgrade?

There is no upgrade in place: uninstall droost and its submodules while their code is still on disk, then install 2.0. Upgrading.

Where is Droost UI Patterns?

It is obsolete. It continues as Droost UI Kit, with the same tools and configuration.

Where do I report a problem?

In the issue queue of the project it belongs to:

  • the module, the recipe, the theme and UI Kit on drupal.org (Droost's queue if you are not sure);
  • the command and the engine on GitHub (droost-workflow).