TokenWatch

AI coding cost attribution for dev agencies. TokenWatch tracks token usage from Cursor, Cline, and Windsurf per developer, project, and client — and surfaces it in a billing dashboard so you can recover AI costs as billable revenue.

What this extension does

When you open a project that has been initialized with TokenWatch, the extension:

Starts a local HTTP proxy on localhost:3030
Automatically points Cursor or Cline's Anthropic base URL to the proxy
Intercepts every AI API call, captures token usage and cost, and forwards it to your TokenWatch dashboard
Shows a status bar indicator with your developer identity

Zero latency impact — usage data is sent fire-and-forget after the API response is already delivered to your tool.

Requirements

Your project must be initialized before the extension can track usage:

npx @tokenwatch/cli init

This creates a .tokenwatch/config.json in your project with your agency's ingest key and your developer identity. The extension discovers this file automatically — no further setup needed.

Team setup: Commit .tokenwatch/config.json to your repo. It contains only the ingest key (no personal credentials). Teammates who clone the repo get automatic tracking on first open — no init needed on their end.

Supported tools

Tool	Status	Notes
Cursor	✅ Supported	Anthropic base URL set automatically
Cline	✅ Supported	API base URL set automatically
Windsurf	🔜 Upcoming	—

Switch tools via Settings → TokenWatch → Tool.

Status bar

The status bar item in the bottom-right corner shows the current state:

Status	Meaning
`⬡ TokenWatch: tracking (you@agency.com)`	Proxy running, usage being captured
`⬡ TokenWatch: not configured`	No `.tokenwatch/config.json` found — run `npx @tokenwatch/cli init`
`⬡ TokenWatch: error`	Proxy failed to start — click to retry

Click the status bar item to restart the proxy.

Commands

Open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P) and search "TokenWatch":

TokenWatch: Restart Proxy — restarts the local proxy
TokenWatch: Stop Proxy — stops tracking for this session
TokenWatch: Show Setup Instructions — shows the proxy URL with a copy button
TokenWatch: Show Debug Info — shows current workspace paths

How it works

Cursor / Cline makes an API call
  → routed through localhost:3030 (this extension's proxy)
  → forwarded transparently to Anthropic / OpenAI
  → response streamed back to your tool with zero added latency
  → token usage parsed from response
  → POSTed to app.tokenwatch.one/api/ingest (fire-and-forget)
  → appears in dashboard under your developer name + project

Project attribution is automatic — the proxy reads your git remote URL and maps it to the correct project in your dashboard, regardless of local directory structure.

Dashboard

View spend by client, project, developer, and branch at app.tokenwatch.one.

TokenWatch is $99/mo per agency. Get started.

Data & Privacy

This extension collects the following data and sends it to the TokenWatch ingest API (app.tokenwatch.one/api/ingest):

Token usage — input, output, and cache tokens per API call
Cost — calculated from model pricing
Developer identity — your name and email from git config user.email
Git metadata — repository remote URL and current branch name
Tool — which AI coding tool made the request (Cursor, Cline)

This data is sent to your agency's TokenWatch dashboard only. It is never sold or shared with third parties.

The local proxy on localhost:3030 is required to intercept API calls before they reach the AI provider. Your API keys pass through the proxy to the upstream provider — they are never stored or logged by TokenWatch.