Usage at a glance, tokens as you type
Token Counter is a lightweight extension for claude.ai that keeps your Claude usage limits and a live token count in view — before you hit them.
Welcome back
Token Counter for Claude
The quota on your plan is real, yet claude.ai never shows it — so people end up opening claude.ai/usage mid-task and guessing at the remaining room. Token Counter closes that gap by splitting every message into three numbers you can act on: Draft prices your typing live, Chat totals the whole conversation, and Overhead covers prompt, features and memory. The whole conversation is re-sent with every turn, so Chat climbs far faster than the words on screen — which is why Claude usage limits arrive sooner than anyone expects.
Everything runs on your device with the open source openai tokenizer as the engine, so each count token check lands in milliseconds and nothing is uploaded anywhere. No API key, no login, no account access — an llm token counter that lives where you type and watches your claude limit for you. Developers pasting long files, writers in all-day sessions, analysts who lean on an ai token counter: anyone pacing their claude usage.
Counts while you type
Draft prices your typing live — a running token count before you hit send.
Whole chat in tokens
Chat totals the entire conversation, which is re-sent with every turn.
Shows invisible overhead
Prompt, features and memory quietly add tokens until you switch them off.
Local token counting
Runs entirely in your browser. Nothing transmitted, nothing collected — ever.
Try the tokenizer
The same open-source gpt tokenizer that powers the extension — a live token calculator for any text.
Powered by the open-source gpt-tokenizer (o200k). Claude tokenizes a little differently, so counts are close estimates — the same “~” you see in the extension. Modern tokenizers are quick — wondering how many tokens is 100 words? Paste 100 words and see for yourself (about 130).
Frequently asked questions
How many tokens is 100 words?
Roughly 130 — simple wording lands lower, technical wording higher. That's why the Draft meter reads your actual text instead of estimating from length.
Is this the token counter OpenAI users already know?
Same engine, new home: the open-source gpt tokenizer behind any openai token counter, put to work for Claude.
Will the anthropic token count match Claude's official numbers?
Closely, but not exactly — the engine is a GPT-family tokenizer, and the claude tokenizer splits text a little differently. Treat every figure as a close estimate, the same “~” the extension shows.
Where do claude ai usage limits actually appear?
Only after you hit them — claude.ai warns you at the lockout, not before. The meter counts continuously, so you see it coming long before the message does.
See your usage before you hit the limit
Turn an invisible quota into numbers you control — a claude token counter on every conversation.
Add to Chrome