Claude HUD: The Plugin That Shows You Exactly What Claude Code Is Doing (and How Much It's Costing You)

You launch a task in Claude Code, press enter, and… wait. The cursor blinks. Something’s happening. But what, exactly? Is it reading files? Running an agent? About to write to disk? Burning through your context window? You have no idea.

That’s the problem Claude HUD solves.


A HUD for Your Terminal

Claude HUD is a Claude Code plugin that adds a persistent status bar at the bottom of your terminal session. Always visible. Always updated. No separate window, no tmux panel, no dashboard to open — it lives exactly where you’re already working.

What it shows, in real time:

  • Context window usage — a progress bar showing what percentage of your context is already consumed
  • Usage quota — how much of your 5-hour or weekly limit you’ve already burned through
  • Active tools — which tools Claude is running right now (Bash, Read, Write, MCP calls)
  • Running agents — which subagents are active in your current session
  • Overall progress — tasks Claude planned vs. tasks already completed
  • Session cost — actual cost in tokens, pulled directly from Claude Code’s native cost data, with local estimation as fallback

It uses Claude Code’s native statusLine API — which means it’s a first-class integration, not a hack. It’s built in TypeScript and the source code is completely open on GitHub.


Why It Matters More Than You’d Think

The context window is one of the most important resources in any Claude Code session, and also one of the most invisible. You can be at 80% without realizing it — and that’s exactly when quality starts to degrade, agents start losing context, and tasks fail in subtle ways.

With Claude HUD, the bar is always there. You watch it fill up. You know when to run /compact before things get messy.

The cost display is equally valuable, though less obvious. Running Claude Code with Opus in a long session can get expensive. Knowing your cost in real time changes how you make decisions: do you spin up three subagents in parallel for this, or is a single focused pass smarter?

For developers in Latin America working with limited token budgets — whether on Pro, Max, or API consumption — that number on your status bar is the difference between intentional use and a surprise at month’s end.


Setup in Under a Minute

Claude HUD installs through Claude Code’s plugin system:

/plugin marketplace add jarrodwatts/claude-hud
/plugin install claude-hud
/claude-hud:setup

The setup wizard walks you through three presets (Full, Essential, Minimal), lets you choose the language for labels, and lets you fine-tune each element individually. After setup, you restart Claude Code and the HUD appears automatically.

Advanced configuration lives in ~/.claude/plugins/claude-hud/config.json — colors, thresholds, layout, time format, and more are directly editable.

One gotcha worth knowing: on Linux, if /tmp is a separate filesystem (tmpfs), the plugin installation can fail. The fix is to run the installation from a session with TMPDIR pointing to a persistent path. It’s documented in the README.


The Numbers

  • 16,600+ stars on GitHub
  • 698 forks
  • 326 commits — actively maintained
  • Listed in the official Composio awesome-claude-plugins registry

For a plugin that does one focused thing — show you what’s happening — those adoption numbers are remarkable. They suggest that many Claude Code users were feeling the same pain: working blind, no idea what Claude was doing or how close to the limit you were running.


Limitations Worth Knowing

The memory usage display (display.showMemoryUsage) is explicitly opt-in and reports approximate system RAM, not precise Claude Code process memory. The README is honest about it: it can overestimate pressure because OS cache and buffers can count as used memory.

Also, Claude HUD reads from Claude Code’s stdin stream and session transcript. It doesn’t have access to information that Claude Code itself doesn’t expose — so if a tool call doesn’t appear in the event stream, the HUD won’t see it either.


The Big Picture

Claude HUD is part of a broader shift in how developers build around Claude Code. The tool itself is powerful, but it was designed for autonomy — it runs, acts, commits. The plugin ecosystem is now filling in the observability layer that was always missing: what’s happening, what did it cost, what changed.

If you’re using Claude Code seriously, you’re moving a lot faster than before. Claude HUD is your instrument panel.

GitHub: github.com/jarrodwatts/claude-hud