Cloken

Tracking since Apr 21, 2026.

Changelog

• May 3, 2026
  description

  Cloken — Context Window Monitor for Claude
  You're mid-conversation with Claude. The responses are getting vaguer. It's forgetting what you said three pages ago. You don't know why — and claude.ai gives you no indication that anything is wrong.
  Cloken fixes that.
  
  What it does
  Cloken adds a small, persistent HUD to claude.ai that shows you a real-time estimate of how much of Claude's context window your conversation is consuming. The context window is the total amount of text Claude can "hold in mind" at once — your messages, its responses, and a chunk of invisible system-level overhead that Claude uses behind the scenes. When that window fills up, quality degrades. Cloken makes that process visible.
  The estimate accounts for:
  
  Every message you've sent in the current conversation
  Every response Claude has given
  A safety margin for system prompts, tool definitions, and other infrastructure Claude uses that you never see
  
  All numbers are labeled as estimates. They're accurate enough to be useful; they're not perfect.
  
  The four zones
  Cloken color-codes your current position in the context window:
  🟢 Green (0–30% of context) — You're in the clear. Claude has plenty of room and is operating at full capacity. No action needed.
  🟡 Yellow (30-60% of context) — You're approaching the midpoint. Most conversations work fine here, but if you're working on something that requires precise instruction-following or recall of early details, you may start to notice subtle drift.
  🟠 Orange (60-85% of context) — Real degradation is likely. Claude is working hard to maintain coherence across a long conversation, and the results often show: instructions get partially forgotten, earlier context gets compressed away, responses feel less precise. This is the zone where most users notice something feels "off" without knowing why. Cloken tells you why.
  🔴 Red (85% or more) — You're deep in the context window. Claude's responses may meaningfully degrade from here. If you're doing anything important — writing, analysis, coding, roleplay with continuity — this is the time to wrap up or start fresh.
  Cloken shows a brief warning the first time you cross into orange and red territory in a given conversation. It won't repeat them mid-conversation if you dismiss them, and it resets cleanly when you open a new conversation.
  
  The HUD
  The context meter lives in the bottom-right corner of your screen as a small, unobtrusive pill — just a compact visualization and a number like "47k / 200k." It stays out of your way while you're writing.
  If it's in the wrong place, drag it anywhere on the screen. It remembers where you put it across page loads.
  If you want it completely out of the way, click the chevron and it will collapse into an unobtrusive tab; still updating, just not taking up space. Click the tab to bring it back.
  Hover over the HUD to expand it for a larger view. When you move your mouse away, it collapses back to the pill automatically.

  Cloken — Context Window Monitor for Claude
  You're mid-conversation with Claude. The responses are getting vaguer. It's forgetting what you said three pages ago. You don't know why — and claude.ai gives you no indication that anything is wrong.
  Cloken fixes that.
  
  What it does
  Cloken adds a small, persistent HUD to claude.ai that shows you a real-time estimate of how much of Claude's context window your conversation is consuming. The context window is the total amount of text Claude can "hold in mind" at once — your messages, its responses, and a chunk of invisible system-level overhead that Claude uses behind the scenes. When that window fills up, quality degrades. Cloken makes that process visible.
  The estimate accounts for:
  
  Every message you've sent in the current conversation
  Every response Claude has given
  All attached documents and images
  MCP Connectors
  
  All numbers are labeled as estimates. They're accurate enough to be useful; they're not perfect.
  
  The four zones
  Cloken color-codes your current position in the context window:
  🟢 Green (0–30% of context) — You're in the clear. Claude has plenty of room and is operating at full capacity. No action needed.
  🟡 Yellow (30-60% of context) — You're approaching the midpoint. Most conversations work fine here, but if you're working on something that requires precise instruction-following or recall of early details, you may start to notice subtle drift.
  🟠 Orange (60-85% of context) — Real degradation is likely. Claude is working hard to maintain coherence across a long conversation, and the results often show: instructions get partially forgotten, earlier context gets compressed away, responses feel less precise. This is the zone where most users notice something feels "off" without knowing why. Cloken tells you why.
  🔴 Red (85% or more) — You're deep in the context window. Claude's responses may meaningfully degrade from here. If you're doing anything important — writing, analysis, coding, roleplay with continuity — this is the time to wrap up or start fresh.
  Cloken shows a brief warning the first time you cross into orange and red territory in a given conversation. It won't repeat them mid-conversation if you dismiss them, and it resets cleanly when you open a new conversation.
  
  The HUD
  The context meter lives in the bottom-right corner of your screen as a small, unobtrusive pill — just a compact visualization and a number like "47k / 200k." It stays out of your way while you're writing.
  If it's in the wrong place, drag it anywhere on the screen. It remembers where you put it across page loads.
  If you want it completely out of the way, click the chevron and it will collapse into an unobtrusive tab; still updating, just not taking up space. Click the tab to bring it back.
  Hover over the HUD to expand it for a larger view. When you move your mouse away, it collapses back to the pill automatically.

Permissions & access

Permissions

storageactiveTabtabs

Host access

https://claude.ai/*

Screenshots