TL;DR: AutoShelf ($19.99 one-time) ships an embedded MCP server that works, but do you actually need that? An AI model can write a launchd LaunchAgent and a shell script once and it runs forever, no app, no daemon, no purchase. The MCP's real value only shows up once you're managing enough rules to want conversational audit and editing instead of hand-maintained scripts, and even there, my hands-on testing found the rule engine underneath thin: no dynamic subfolder sorting, limited management by file size, and an action list a fraction of Hazel's. AutoShelf delivers less using AI than its marketing promises and loses the file-organizer argument outright.

AutoShelf

I've had a Hazel rule watching my Downloads folder since before I can accurately date it. It reads PDF invoices and files them by vendor, decompresses ZIPs and moves the app inside to Applications, converts HEICs to JPG, and hands ebooks to Calibre and audiobooks to Audiobookshelf without me touching any of it. That's the bar. A new license for Hazel is $42, which is steep for some and I've written before about cheaper options. AutoShelf wants to be added to that list, and it's making its case with a feature none of those alternatives had: an MCP server for use with AI agents plus a CLI.

More Than Just Simple File Management

• An MCP server built into the app binary, not a plugin bolted on after the fact. Point Claude Desktop, Cursor, Codex, or Windsurf at it and describe a rule in plain English instead of building it in a UI.
• A companion CLI for scripting rule creation and running jobs from the terminal, the same engine the MCP server talks to.
• Where-from metadata as a condition, so a rule can act on which browser or app downloaded a file, not just its type or name.
• OCR and Live Text matching as of version 2.2, so a rule can file a scanned invoice by what's written on it. Not new ground: Hazel's Contents attribute has done on-the-fly text recognition on images and text-layer-less PDFs for years, it's the mechanism behind the vendor-filing rule in the intro. AutoShelf catching up to a feature Hazel already had is parity, not an edge.
• Multi-action, chainable rules: move, tag, rename, and archive to ZIP in a single pass, added in the 2.0 rewrite.

Is It Better Than Hazel for Working With AI?

Does anyone even need a layer between a prompt for file management and their file system? For a well-defined one-off job, not really. A capable AI model can write a launchd LaunchAgent with a WatchPaths key, native FSEvents, no polling loop, and a small shell script that switches on extension and moves the file. Write it once and it runs forever: no app, no background daemon, no $19.99. That's possible without Hazel or AutoShelf.

An MCP connection can do something a bespoke script can't: with it running, AutoShelf can immediately edit any rule with natural language, without you having to track down the launchd job and script you created six months ago. I created eight AutoShelf rules and asked Claude to check on them through the MCP connection; it returned current rules, current paths, and current enabled/disabled flags. A launchd script has no equivalent self-report; ask it "what are you doing" six months from now and the honest answer is "go read the shell and simulate it," and any edit needs a launchctl unload/load before it's live. MCP matters once you're managing enough rules that you'd rather audit, pause, or edit them by asking than by maintaining scripts by hand, or you want the GUI, activity log, and undo the app provides that a bare launchd job doesn't have unless you build it yourself.

Against Hazel specifically, AutoShelf is still ahead on this one front. Hazel doesn't have an MCP server and doesn't claim to; anything an AI does to a Hazel rule today runs through a general-purpose file-editing layer poking at Hazel's XML underneath, which is fragile and unsupported. So among app-based rule engines with a GUI and an activity log, AutoShelf's MCP integration is real. On my Mac, it was easy to set up and worked as advertised on the first try.

The Rule Engine Underneath Doesn't Hold Up

This is where my hands-on testing exposed the weaknesses in AutoShelf. It can't sort a folder of mixed files into subfolders by extension unless you create every destination subfolder yourself first and then write a separate rule for each file type. I had to build five nearly identical rules, "Sort PNG files," "Sort JPG files," "Sort JPEG files," "Sort GIF files," "Sort MD files," each one hand-built with its own hardcoded destination path. Hazel does the same job with a single "Sort into subfolder" action and a pattern; AutoShelf makes you write and maintain N rules for N extensions.

Two more gaps from my testing: size conditions only go down to whole megabytes, no decimal places, so you can't write a rule for anything under 1 MB or between whole-MB steps. More importantly, AutoShelf's action list (Move, Copy, Trash, Rename, Add Tag, Archive to ZIP, Optimize, Convert Image/Video/Audio, Photos import) is a fraction of Hazel's, which adds sync, upload, tags, color labels, comments, alias creation, three Music/Photos/TV import targets, and five different scripting hooks (Shortcuts, AppleScript, JavaScript, Automator, shell) on top of pause, notify, and rule-chaining controls.

The Recursion Illusion

AutoShelf's App Store listing says it will "watch any folder you choose, including nested subfolders," and the 1.5 changelog calls it "Nested Folder Watching." (App Store listing) But on its own Product Hunt launch, the developer told a commenter directly that watched folders are top-level only, files dropped into a subfolder are not picked up, and true recursion is a tracked feature request, not a shipped one. (Product Hunt) Read charitably, "nested folder watching" means you can manually add specific subfolders you already know about as additional watch targets, one at a time. That is not the same as what Hazel does, which is recursive: add one "Run rules on folder contents" rule and Hazel will walk into subfolders of subfolders it has never seen, at any depth, applying your existing rules the whole way down. (Noodlesoft manual) That's a real discrepancy between AutoShelf's marketing copy and its confirmed current behavior, and now my own rule list shows the practical cost of it: I'm hand-maintaining what Hazel would do with one rule.

Who This Is For, and Who Should Stay on Hazel

If your files land flat, one folder, no meaningful variety in file type, and you are considering a $19.99 one-time tool with a working AI hook, AutoShelf's free tier is worth an afternoon. If your needs are more complex, stick with Hazel. My own eight-rule test folder is the case study: if you have more than three or four file types, or anywhere the folder structure has depth, AutoShelf turns into more manual rule-building than the automation it's supposed to replace.