Agent HUD
Supported agents

Claude Code: quota, approvals and replies

Claude Code gets everything Agent HUD offers: its plan limits at the edge of your screen, its requests answered on the HUD, and every turn's tokens in your statistics.

Updated 2026-10-29

At a glance

FeatureClaude Code
Tokens and sessions✓
Quota✓
Live status✓
Approvals✓
Questions✓
Replies✓
One click back✓
Turns and calls✓

Compare all supported agents

What you see on the HUD

The most used Claude plan limit sets Claude Code's colour in the glow. The hover lists the session limit, the weekly limit across models and each model family's weekly limit, with their resets or burn rates, and Claude Code's sessions, those waiting for you first.

When Claude Code asks to run a tool, you allow it once, deny it, or always allow it with the rule Claude Code suggests. Its questions and the plans it asks you to approve are answered on the HUD too, the questions Claude asks in its desktop app included. A request you leave goes back to Claude Code after 10 minutes.

Which account does it count against?

The account Claude Code is signed in to, labelled with its email and plan; a Max plan shows its 5× or 20× tier. A reading taken while the sign-in changes is dropped, never mixed. An API key or a third-party platform has no plan limits, so no quota appears; tokens and sessions still count.

Running turns and replies

The HUD shows Claude Code working, waiting for you or done; in Dynamic mode its logo becomes Clawd while it works. A finished turn brings a New reply reminder, and one click takes you back to the session: in the Claude desktop app, or in its iTerm pane when it runs outside tmux. Each session's page lists its turns and calls: the new tokens of each turn, and each call's model, context and tools. Sub-agents count inside the session that started them.

Requests and replies arrive through client hooks, which the release writes into Claude Code's own settings while they are on and removes when they are off. A build from the source on GitHub does not install them yet.

What does Agent HUD read?

The transcripts Claude Code keeps in its projects folder, for tokens, turns and sessions. For the plan limits it asks the installed Claude Code itself, with its hooks off. The reading sends no message to a model, and no credential is stored.

On iPhone, Apple Watch and with other agents

On the same network as the release, iPhone and Apple Watch show Claude Code's quota, tokens and sessions without a subscription. Agent HUD Remote adds the rest, anywhere: read its conversations, send it messages that arrive as if you had typed them, answer its requests, and turn a note spoken to your watch into a new session.

Through the built-in MCP server in the release, Claude Code can ask a Codex session to review a change and get the answer back.

Questions

Does reading Claude Code's quota use any of it?

No. Agent HUD asks the Claude Code you already signed in to for its usage figures. No message goes to a model.