@kunal_dutt/token-lens 0.2.33 → 0.2.34

package/PRIVACY.md

@@ -18,7 +18,7 @@ anyone.
 | --- | --- | --- | --- |
 | **Browser extension** (Chrome, Edge, Firefox) | The prompt box, model picker and visible messages on the AI chat sites listed in its permissions, to count tokens. | Your settings, prices, budget and savings totals, in your browser's own extension storage. Uninstalling deletes them. | Nothing. Its only network request is on claude.ai: at most once a minute it reads your own session and weekly usage from claude.ai, signed in as you, so plan tips use real numbers. The result is not stored and goes nowhere else. |
 | **MCP server on your computer** (`npx @kunal_dutt/token-lens`, Claude Desktop extension) | Only the text an AI app passes to its tools. | Savings totals only: per day, action and model, the number of tokens and dollars saved, in `~/.token-lens/savings.json` on your computer. Never any text. Delete the file to clear it; `TOKEN_LENS_LEDGER=off` turns it off. | Nothing by default. Only if you set your own `ANTHROPIC_API_KEY` or `GEMINI_API_KEY` for it does it send the text to that provider's token-counting endpoint, with your key, for an exact count. |
-| **Live tips in the Claude app on a Mac** (part of the MCP server when the Claude desktop app starts it) | Through macOS Accessibility, once you allow it: the text in Claude's message box and the open chat, to count tokens. | Only on/off settings in `~/.token-lens/live.json`. The text stays in memory while counting and is never written down. | Nothing. macOS notifications on your Mac, and a panel served only to your own Mac at `localhost:4797` that refuses other sites. |
+| **Live tips in the Claude app on a Mac** (part of the MCP server when the Claude desktop app starts it) | Through macOS Accessibility, granted to Token Lens Helper once you set it up: the text in the front AI app's message box and open chat (Claude or ChatGPT), to count tokens. | Only on/off settings in `~/.token-lens/live.json`, and the helper app itself in `~/.token-lens/helper`. The text stays in memory while counting and is never written down. | Nothing. macOS notifications on your Mac, and a panel served only to your own Mac at `localhost:4797` that refuses other sites. |
 | **Your own connector** (`npx @kunal_dutt/token-lens --http`) | Only the text sent to its tools. | Nothing. | Nothing beyond the reply. |
 | **Public hosted connector** (`/api/mcp`) | Only the text an AI app passes to its tools. | Nothing. Each request is handled in memory and discarded; requests are not logged or stored. The hosting provider keeps standard request metadata (time, IP address, URL) for security, as any website does. | Nothing beyond the reply. Prefer running your own copy if even that metadata matters. |
 | **Developer library** (`@kunal_dutt/token-lens/sdk` on npm, `token-lens` on PyPI, inside your app) | Your app's requests to OpenAI or Anthropic, in your own process. | Savings counters in your app's memory. | Nothing of its own. Your requests still go only to the provider your code already calls. |

package/README.md

@@ -50,9 +50,14 @@ Then ask, for example: "optimize this prompt with Token Lens", "how full is this
 chat?" or "optimize this agent's system prompt and tools".
 
 **Live tips in the Claude app on a Mac, with no command.** When the Claude
-desktop app starts Token Lens, it also watches Claude's message box and the open
-chat (through macOS Accessibility, after you turn on Claude once in System
-Settings > Privacy & Security > Accessibility). When you pause typing and a
+desktop app starts Token Lens, it also watches the message box and open chat of
+the AI app in front (Claude, or ChatGPT's Mac app), through macOS Accessibility.
+That permission goes to **Token Lens Helper**, a small app Token Lens builds in
+`~/.token-lens/helper` only when you click **Set up Token Lens Helper** at
+<http://localhost:4797> (or ask Claude to "set up Token Lens Helper"), then turn
+on in System Settings > Privacy & Security > Accessibility. The helper holds
+the permission for Token Lens alone, with its watcher compiled in, so you never
+need to turn on "node", which would give every Node program that control. When you pause typing and a
 shorter prompt says the same thing, or a chat gets long enough that every
 message re-reads a lot, a macOS notification says so, in tokens and dollars.
 The shorter draft waits, ready to copy, at <http://localhost:4797>. It never

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@kunal_dutt/token-lens",
-  "version": "0.2.33",
+  "version": "0.2.34",
   "description": "Count and cut the tokens your prompts, chats and agents use, adapted to the model and task. MCP server and self-hosted gateway. Collects no data.",
   "keywords": [
     "mcp",