The first time I pointed Claude Code at a new project, I did what most people do: almost nothing. I typed my question, got an answer, and left every setting where it was. I knew the thing could run hooks, dispatch subagents and connect to external servers, but configuring all of that felt like an evening of research I never got around to. The project worked, so why bother.
That is exactly the gap Anthropic is now trying to close with an official plugin: claude-code-setup. It does one thing, and it does it well. It looks at your project and says: given what is here, I would turn on these few things. Not everything that is possible, but the one or two that pay off most for your particular codebase.
What it is
Under the hood sits a skill called claude-automation-recommender. The official description is dry and precise: “Analyze codebases and recommend tailored Claude Code automations such as hooks, skills, MCP servers, and subagents.” You enable the plugin, you say something like “recommend automations for this project” or “what hooks should I use?”, and Claude goes to work.
What it does then is read your project. It checks whether there is a package.json or a pyproject.toml, which frameworks you use, whether a .claude folder and a CLAUDE.md already exist, what your directories are named. From that it suggests what fits, spread across five kinds:
- Hooks for automatic actions, like formatting or linting on every change, or blocking edits to sensitive files.
- Skills for packaged expertise and repeatable tasks.
- MCP servers for connections to the outside world, such as a browser or a documentation source.
- Subagents for specialized reviewers that watch in parallel for security or accessibility issues.
- Slash commands for quick, fixed workflows.
Per kind it deliberately keeps it short: the top one or two, not a long list you will not read anyway. Ask explicitly about a single category and it offers more. And it does not stop at a fixed list; for the tools and libraries specific to your project it searches the web for what suits them. The plugin comes from Anthropic itself, written by Isabella He, and lives in the official plugin repository that by now has well over thirty thousand stars.
The detail I liked
One line in the documentation stood out. The skill is explicitly read-only. It analyzes and advises, but creates or changes no files of its own. Whatever it recommends, you build afterwards, or you ask Claude in a separate step to do it for you.
That sounds like a limitation, but it is a design choice I appreciate. There is a clear line between advice and action. The thing that makes the recommendations is not the same thing that carries them out. You get a list, you read it, and you decide what ends up in your project. In a time when tools more and more often arrange things for you unasked, an adviser that explicitly keeps its hands to itself is a welcome exception.
The caveat that comes with it
And still. A recommendation engine always pushes in one direction, and here that direction is: more. More hooks, more subagents, and above all more MCP servers. Every MCP server you switch on is a connection to something outside your own machine, and every connection is a bit of dependency you take on. That is no objection to the plugin, since it keeps things tidy at one or two per kind. But the sum across several projects is something to watch.
My own rule is simple: a recommendation is a question, not an order. Before I turn on a server or a hook I want to be able to explain what it does, where its data goes, and what breaks if it disappears tomorrow. For most suggestions a recommender like this makes, that answer is fine. For a few it is reason to say a polite no. The plugin takes the legwork off your hands; the judgment you keep for yourself.
What I end up liking is that this lowers the threshold without taking away control. For anyone using Claude Code the way I did at first, with every setting on default, it is a good way to see what you are leaving on the table. You do not have to follow the advice. But at least now you know what there is to choose from.
Sources
- claude-code-setup plugin — Anthropic, official plugin repository
- claude-automation-recommender skill — source skill with the full workflow