3 Open-Source Tools We Built for Claude Code (And Why We Released Them)

Displace Agency open-sourced three developer tools built for Claude Code, Anthropic's AI coding assistant: a statusline configuration agent that sets up terminal status displays, an MCP server launcher that manages Model Context Protocol connections, and a CPU watchdog that monitors resource usage during AI-assisted development sessions. All three are available on GitHub as free, MIT-licensed packages.

These three tools solve problems we hit daily: not knowing how much a conversation costs, overloading the system with unnecessary MCP servers, and catching runaway processes before they melt the laptop. Each took less than a day to build. Each saves time every single session.

The default Claude Code statusline shows the model name and not much else. Our enhanced statusline displays real-time token usage with K/M formatting, API costs in dollars, cache hit rate percentage, message count, precise session duration, and remaining context. All updated on every message.

The practical impact: you know exactly when a conversation is getting expensive, when cache efficiency is dropping, and when you're approaching the context limit. No more guessing whether to continue a thread or start fresh. The statusline tells you. In our daily usage, we've seen sessions range from $0.50 for quick fixes to $15+ for full-page builds. Without the statusline, you're flying blind.

• Real-time token count with K/M formatting (e.g., 153.5K tokens)
• API cost tracking in dollars (e.g., $4.11 this session)
• Cache efficiency percentage (e.g., 92% cached)
• Context remaining indicator (know exactly when to start fresh)
• Session duration with seconds precision
• Git branch and sync status integrated

# Install Claude Statusline Enhanced
git clone https://github.com/displace-agency/tool-claude-statusline-enhanced
cp tool-claude-statusline-enhanced/statusline.sh ~/.claude/
# Add to ~/.claude/settings.json:
#   "statusLine": "~/.claude/statusline.sh"

MCP (Model Context Protocol) servers give Claude Code access to external tools like Google Workspace, Figma, Notion, and hundreds more. The problem: every MCP server loaded at startup consumes tokens for tool definitions, even when unused. With 8+ MCP servers configured, we measured 12,000-15,000 tokens consumed at startup just loading tool schemas. That's $0.15-0.20 per conversation before you type a single word.

Our MCP Launcher provides three commands: 'claude' loads only Rube (the most token-efficient server, on-demand fetching), 'claude-all' loads everything, and 'claude-custom' shows an interactive menu to pick exactly which servers you need. When Claude exits, the original config is automatically restored. In daily use, we run 15-20 Claude sessions. At $0.15 saved per session, that's $2-3/day or $60-90/month in token savings.

• Three launch modes: minimal, full, and custom selection
• Auto-restore of original config on exit (even on crash)
• Interactive terminal menu for custom selection
• Rube always active (500+ app integrations, tokens only when used)
• Saves ~12,000 tokens per session when loading minimal config
• Zero dependencies beyond jq

# Three modes for any context
claude          # minimal ,  Rube only
claude-all      # everything loaded
claude-custom   # interactive picker

Running multiple Claude Code sessions simultaneously caused Terminal to spike to 722% CPU and the Mac to 101 degrees Celsius for over an hour. That experience led directly to CPU Watchdog: a lightweight launchd daemon that monitors process CPU usage every 30 seconds and sends native macOS notifications when something runs hot for too long.

The design is deliberately conservative. It waits for 2 consecutive minutes above 200% CPU (equivalent to 2 full cores) before alerting, safelists system processes like kernel_task and WindowServer, and re-alerts every 2 minutes while the process stays hot. The script is 47 lines of bash. No dependencies. It has caught 3 runaway processes since we installed it, including a forgotten Next.js build that was consuming 400% CPU for 20 minutes.

We release these tools because the Claude Code ecosystem is young and growing. The developers building the next generation of AI-assisted software need practical solutions, not theoretical frameworks. Every tool we release under MIT license is a potential conversation with someone who might become a client, a collaborator, or a contributor.

Open-sourcing developer tools is also honest marketing. It shows what we build, how we think, and the quality standard we apply even to internal utilities. If our throwaway shell scripts are well-documented and production-ready, imagine what we deliver for paying clients.

• All three tools available at github.com/displace-agency
• MIT licensed, free for any use
• Each solves a real problem we hit in production
• Contributions welcome via pull requests