5 Tools I Built for Claude Code – And Why I Use Them Every Day

I work with Claude Code every day. Not as a toy — as a tool that generates production code. (More about me and my work with AI.) And when you use something this intensively, you quickly hit edges that slow you down.

Instead of complaining, I wrote my own solutions. Here they are.

1. claude-gc — Garbage Collector for Claude Code

The problem: Claude Code spawns child processes — subagents, MCP servers, Task workers. When you close the terminal, these processes keep running. Each one eats ~220 MB of RAM. Within hours, you can have hundreds of orphans consuming your server.

This isn’t theoretical. I’ve seen 1,000+ orphaned processes consuming 75 GB on a 32 GB server. OOM kill. Server dead.

One-line install:

curl -fsSL https://raw.githubusercontent.com/kojott/claude-gc/main/install.sh | bash

Works on Linux, macOS, and WSL. Zero dependencies — pure Bash.

→ GitHub: kojott/claude-gc

2. claude-notify-telegram — Telegram Notifications

The problem: You start Claude Code on a long task, walk away — and have no idea when it’s done or if it’s waiting for your input.

This plugin hooks into Claude Code events (Stop and Notification) and sends a message via the Telegram Bot API. When Claude finishes work or needs your permission — you get a push notification on your phone.

Silent failure — if credentials aren’t configured, the plugin does nothing. Only requires jq and curl.

→ GitHub: kojott/claude-notify-telegram

3. claude-docu-optimizer — Documentation Optimizer

The problem: The CLAUDE.md file is persistent context for Claude Code. Over time it bloats, goes stale, stops matching reality. Claude then works with inaccurate instructions — and you don’t know why.

This plugin is based on best practices from Boris Cherny’s team at Anthropic. It analyzes your CLAUDE.md, detects 10 types of anti-patterns, and suggests optimizations.

8 modes: analyze, optimize, apply, compare, create, sync, audit, scaffold.

→ GitHub: kojott/claude-docu-optimizer

4. remoteclaud — Remote Server in One Command

The problem: You want Claude Code on a remote server, but setup takes hours. Docker, Node.js, Go, Python, tmux, Claude CLI — and it all needs to work together.

An Ansible playbook that turns a bare Rocky Linux ARM64 VPS into a complete Claude Code development environment. One command.

What it installs: Docker, Node.js, Python, Go, Claude CLI, and Claunch — an interactive tmux session manager for Claude. Type cl and you get a menu with your Claude sessions.

ansible-playbook -i inventory.ini playbook.yml

Perfect for Hetzner Cloud ARM64 instances — performance at a reasonable price.

→ GitHub: kojott/remoteclaud

5. DropShot — Menu Bar Upload for Developers

DropShot is a native macOS menu bar app that solves a simple problem: you need to send a file to a server and get the path in your clipboard.

Drop a file on the menu bar icon → SFTP upload → absolute path in clipboard. Or Cmd+Shift+U for a screenshot with instant upload.

Why? Because in tmux/terminal sessions there’s no drag & drop. And re-sharing files through cloud services with expiring links is a waste of time.

Swift 5.9, SwiftUI, Keychain integration, SSH key support including 1Password. The full story of how I built it in an afternoon from my phone is in a separate article.

→ GitHub: kojott/dropshot

Why I Do This

Every one of these projects started from a real need. None were planned as “open-source products.” I was solving a problem that slowed me down — and then realized everyone who works with Claude Code intensively has the same problem.

All tools are free, open-source, and ready to use today. And if you want a practical overview of commands and techniques for working with AI, grab the free AI Cheat Sheet for Developers.

If you’re curious about how I work with AI tools and how they can transform your team’s productivity — get in touch.

You might also like

• DropShot: a macOS app in one afternoon
• Why 90% of AI trainings end up changing nothing
• How much does a day without AI cost you?