Claude File Token Counter

Count exactly how many tokens the current file (or selection) costs against a specific Claude model — on demand, from the status bar.

Why

Token counts are model-specific — every Claude model has its own tokenizer. This extension asks the official Claude token-counting endpoint (POST /v1/messages/count_tokens) for the exact input_tokens of your file against the model you choose. No estimates.

It does not use tiktoken — that is OpenAI's tokenizer and miscounts Claude text by 15–20%+.

Features

• 📊 Exact counts from the official Claude endpoint, not an estimate.
• 🖱️ On demand — one click in the status bar; nothing runs automatically.
• ✂️ Selection or whole file — counts a non-empty selection (Selected N tokens) or the entire document (N tokens), including unsaved edits.
• 🧠 Model-specific — set the model, or detect it from your Claude Code usage.
• 🔐 Key stays secret — stored in VS Code SecretStorage, never in settings or git.
• 🪶 Stays honest — the count clears when you switch editors or edit text, so a stale number never lingers.

Getting started

1. Install

In VS Code: open Extensions (Ctrl/Cmd+Shift+X), search “Claude File Token Counter”, and click Install — or run:

code --install-extension offreal.claude-file-tokens

2. Add your Claude API key

1. Get a key from the Claude Console → API keys: https://platform.claude.com/settings/workspaces/default/keys (click Create Key, copy the sk-ant-… value).
2. In VS Code open the Command Palette (Ctrl/Cmd+Shift+P) and run “Claude Tokens: Set API Key”, then paste the key.

The key is saved in VS Code's encrypted SecretStorage — it is never written to settings.json and never committed to git.

3. (Optional) Choose the model

Set claudeTokens.model in Settings (default claude-opus-4-8), or run “Claude Tokens: Detect Model from Claude Code” to pick a model from your ~/.claude.json usage history for the current project.

4. Count

Click the Tokens item in the status bar. While the request runs it shows Counting…; on success it shows the count. Hover for a tooltip with the model, token count, character count, mode, and file name.

Usage

• Whole file — with no selection, the entire document is counted → N tokens.
• Selection — with a non-empty selection, only that text is counted → Selected N tokens.
• Always uses the current in-memory text, so unsaved edits are included.
• The status bar resets to a neutral Tokens label when you switch editors or edit the text.

Commands

Settings

What the number means

It is the input_tokens of a request where the file content is wrapped in messages: [{ role: "user", content: <text> }]. That includes a small wrapper overhead, which honestly reflects how much actually goes to the model.

Notes

• The Claude SDK runs in the extension's Node host and picks up HTTPS_PROXY from the environment, so corporate proxies work without extra configuration.
• Your file content is sent to the Claude API only when you click to count. Nothing is sent automatically.

Build from source

pnpm install
pnpm run compile      # bundle with esbuild
pnpm run package      # build a .vsix

Press F5 in VS Code to launch an Extension Development Host with the extension loaded.

Lint and formatting:

pnpm run lint         # ESLint
pnpm run format       # Prettier --write

License

MIT