[FEATURE] Claude desktop app chat (not Code): on-demand context usage query, readable by the user and the session

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

In Claude desktop chat there is no way to check context usage when I want to. Claude Code CLI has /context; desktop chat has nothing equivalent.

• The auto-compaction notice shows no numbers.
• Asking the model doesn't help, because it cannot see its own context fill or the limit.

In long, multi-step sessions (dozens of document reads and edits per batch), I need to decide at natural break points whether to continue, write a handoff, or start a new chat. Today that decision is guesswork, and compaction often lands mid-batch.

Proposed Solution

Expose the current context usage (used / limit / %) in desktop chat on demand:

1. User side: a command or menu item (e.g. /context) that shows the figures for the current chat at that moment. No persistent UI needed.
2. Session side: make the same figures readable by the model, for example through a built-in tool. The session can then act on them, e.g. "warn me at 80% and propose a handoff before compaction". I only need to state the threshold in chat.

This is about querying on demand. It is not a persistent indicator, which #46897 and #20041 already request.

Alternative Solutions

• Asking the model to estimate: it can't see the figures.
• Inferring from the compaction notice: it carries no numbers, and compaction can't be partially triggered just to read them.
• Starting new chats preemptively: this throws away context early and is guesswork.
• CLI /context and the status line: not available in desktop chat.

Priority

Medium - Would be very helpful

Feature Category

Other

Use Case Example

1. I start a long review session in desktop chat and tell Claude: "If context passes 80%, finish the current batch and write a handoff."
2. Claude works through several batches, reading and updating about ten documents each.
3. Between batches, I run /context and see "142k / 200k (71%)", so I start one more batch.
4. Mid-batch, Claude reads 81% and finishes that batch. It then gives me a handoff prompt instead of being cut off by auto-compaction.

Additional Context

Checked existing requests. The same idea has been raised for Claude Code, but not for desktop chat:

• #20750 Native context awareness for Claude and users (CLI, closed as duplicate)
• #34184 Expose context usage percentage to the model (Claude Code, closed as not planned)
• #46897 Context usage indicator (Claude Code Desktop app, persistent display, closed as not planned)
• also #16510, #32811, #16255 (exposing usage to the model in Claude Code)

Claude Code at least has /context and the status line. Desktop chat has no equivalent at all, and long agentic chat sessions there are compacted without any figure being available to the user or the model.