Free · open source · runs locally

See everything you’ve installed in Claude — and tidy up the ones you forgot

A free tool that maps every skill, plugin, MCP server, and agent in your Claude setup — split by global vs. project, with real usage counts — so you can clean out the clutter in minutes.

Run this once in your terminal
curl -fsSL https://stackcleaner.com/scan.mjs | node
Writes stack-cleaner.json locally. Secrets stripped. Nothing leaves your machine.
The problem

Your Claude setup grows faster than you can keep track of it

Every install adds up — a skill here, a plugin there, an MCP server you tried once, an agent you copied from a tutorial. They spread across your global config and every project you’ve touched. Months later you can’t answer three simple questions:

“What’s even installed?”

Skills, plugins, MCP servers, and agents are scattered across global and project folders. There’s no single place to see them all.

“Do I actually use this?”

Half of it is from a tutorial you followed once. Without usage counts, you can’t tell what’s earning its keep from what’s dead weight.

“What’s just noise?”

Duplicates, abandoned experiments, and overlapping tools quietly bloat your context and clutter every prompt — and you forget they exist.

How it works

Three steps, about two minutes

No account, no upload, no install to babysit. Scan, look, and tidy.

  1. 1

    Scan

    One line in your terminal walks every .claude folder — global and per-project — and writes a single stack-cleaner.json. Secrets are stripped before anything is written.

  2. 2

    See

    Drop that file into the web app and everything snaps into view — grouped by global vs. project, with real usage counts so you know what’s used and what’s collecting dust.

  3. 3

    Tidy

    Tick the things you want gone and get a ready-to-run cleanup — a shell script, a paste-to-Claude prompt, or plain JSON. Remove the clutter your way, in seconds.

Why it’s safe

Built to be useful and trustworthy

A clean look at your setup, without asking you to trust a black box.

Global vs. project, side by side

Everything is split by where it lives: your global ~/.claude setup and each repo's own .claude, so you always know what's yours everywhere vs. scoped to one project.

Real usage counts

See what's actively used versus what hasn't run in ages. Sort the wheat from the chaff with numbers, not guesswork.

Secrets stripped at the source

Known secret fields (API keys, tokens, MCP env values, and URL credentials) are redacted before the file is written, and your home path becomes ~. The scan never sends anything anywhere.

100% local, nothing uploaded

The scan runs on your machine and the web app parses the file in your browser. No server, no account, no telemetry. Your setup stays yours.

Free & MIT open source

No paywall, no trial, no catch. The whole thing is open source under MIT. Read the code, fork it, or run it yourself.

Hand the cleanup to Claude

Prefer to delegate? Export a paste-to-Claude prompt and let Claude do the removal for you, step by step. The tool only generates commands. It never deletes anything itself.

No terminal experience needed

Never opened a terminal? You’ll be fine.

The setup page walks you through it copy-paste by copy-paste — what to open, exactly what to type, and what you’ll see back. No code to read, no GitHub account, nothing to install and configure. If you can copy a line and press Enter, you can do this.

Ready when you are

See what’s really in your Claude setup

Poke around the demo with no setup at all, or scan your own machine and start tidying. Either way it’s free, local, and takes about two minutes.