Cursor Cost Tracker Install

Version 1.0.8 · Live in Cursor · Free

See the cost — before the month runs out

Cursor spend and local Codex, in one panel. No extra app. No pasted token.

  • Status bar — Current, Today, and recent queries. Green on pace; red on a spike (!).
  • Guards — Burn Rate toasts and a blocking critical alert. Neither stops Cursor.
  • Monthly forecast — green on plan, red above it, and a straight answer: it lasts, or it runs out.
  • Optimize — a ready prompt for the last red query. Never reads the chat.
  • Coding stats — landed / AI = % effectiveness, plus All on Cursor by project.
  • Codex local — the same dates as Cursor, one row per thread, cache hit, and an API estimate that is not an invoice.
1,522 installs on Open VSX
Local session stays on this machine
Free nothing to paste, MIT

Open VSX snapshot

Why this extension

Why Cursor Cost Tracker?

Cursor’s dashboard shows totals on the website — not how much of the month or day is left, when limits run out, which query spiked, or a local way to shrink the next expensive turn.

This extension keeps those numbers next to Git and Problems, with a monthly forecast and Optimize savings in .ai/optimize-savings.md — no chat upload, no third-party analytics.

Full comparison Cursor Cost Tracker vs Cursor Dashboard by job
Cursor Dashboard vs Cursor Cost Tracker, grouped by job
Capability Cursor Dashboard Cursor Cost Tracker
Status bar
Current cycle spend on the bar × Yes
Today vs remaining daily budget on the bar × Yes
1–10 recent queries on the bar × Yes
Green on pace · red on a spike (!, default 1M) × Yes
Guards
Burn Rate Guard (live $/window, banner, non-blocking toasts) × Yes
Blocking critical alert (default 10M tokens or $5) × Yes
Neither guard stops Cursor × Yes
Ignore a spike ! and keep it dismissed after reload × Planned
Monthly forecast
Used / forecast / ideal on one chart × Yes
Run-out date (Team $ or Pro included limits) × Yes
Pace by working days or all calendar days × Yes
Today / 7 days / Month range on the chart × Yes
Optimize
Ready prompt for the last red (Warn at) query × Yes
Quick / Balanced / Deep depths × Yes
Projected save in local .ai/optimize-savings.md × Yes
Never reads the chat transcript × Yes
Coding stats
Landed / AI = % effectiveness × Yes
If this branch landed on main / master × Yes
All on Cursor · split by project × Yes
AI lines vs merged on Charts × Yes
Panel · queries · charts
Last N queries in the editor (100–10,000 or From date) × Yes
Per-query cost, tokens, model, kind Website Yes
Statistics (totals, averages, spike count) Website Yes
Spend by model and by kind Website Yes
Billed spend donut for the loaded sample × Yes
Charts (tokens / cost over time) × Yes
Today / This month / All time mix cards Website Yes
Model pricing (docs + Last N counts + CursorBench) Website Yes
Local Codex in the same date window (API estimate, not an invoice) × Yes
Quota gauge, on-demand, and accepted lines × Yes
Export recent queries as CSV × Yes
Support → Write a message to the author × Yes
Setup · language
UI language (11 locales) × Yes
Zero setup (local Cursor session) × Yes
No token pasted into Settings × Yes

Not in scope: payments, hosted team dashboards, other IDEs, estimating cost before you send a prompt, reading chat transcripts, or auto-rewriting your codebase. Optimize only builds metadata prompts you paste yourself; savings numbers come from the agent writing a local project file.

Inside the editor

It works before you change a setting.

Current and Today land on the status bar the moment you install. Statistics shows the quota, accepted lines, a billed-spend donut, and whether the month lasts. Codex local uses the same dates.

Status bar

Spend stays visible while you type

Current is the billing cycle. Today is the daily pace. Up to ten recent queries sit beside them. Anything at or above Warn at (default 1M tokens) gets a red !.

Statistics

Where the billed dollars went

One donut of the loaded sample. Switch model, vendor, size, limit, kind, conversations, grade, or cache.

Statistics

Quota, lines, and the pace

Included Auto on a gauge, accepted lines for today and the range, a billed-spend donut, and a forecast that stays green on plan.

Queries

Cursor or Codex, one row per chat

Switch source without leaving the editor. Group a thread, see how full the window is, and export a flat CSV.

Optimize

A prompt for the last red query

Quick, Balanced, or Deep. Run pastes into the open Agent chat — you press Start. The extension never reads the transcript.

  • Critical alert. One dialog, only at 10M tokens or $5. Separate from the ! on the bar. Cursor keeps running.
  • Eleven languages. English by default. Polish, Chinese, Japanese, and more switch the panel, status bar, and toasts immediately.
  • Codex stays on this computer. Sessions are read locally. Files are not uploaded. Dollars are an API estimate, not an invoice.
  • Billed spend. One donut of the loaded sample. Switch Model, Vendor, Size, Limit, Kind, Conversations, Grade, or Cache. On Limit, see what Optimize saved.

Product tour

1.0.8, frame by frame.

The newest release is selected first: Codex, billed spend, the cost sample, quota, and the forecast. Older releases stay one chip away. Click a frame for full screen — click again for actual size.

High burn rate — banner and live $/window when the window is hot.
Monthly forecast — used, forecast, and ideal on one chart.
Status bar — Current, Today, and the latest queries.
Status bar cards — burn rate, Current, Today, Coding stats.
Statistics — Current, Today, burn rate, and coding stats.
Billing cycle — plan, cycle dates, and Last N summary cards.
Spend breakdown — by model, by kind, busiest day, token mix.
Forecast near run-out — tip shows used vs allowance for a day.
Last N · spikes — newest first, Warn-at ! on TOKENS.
Last N queries — cost, tokens, input / output, kind.
Optimize — ready prompt for the last red query.
Coding stats — AI lines vs what landed on the default branch.
Model pricing — docs rates, Last N counts, CursorBench.
Support — coffee tip and write a message to the author.

1 / 14

Install in under a minute

No second login. Nothing to paste.

Cursor’s Extensions view searches Open VSX. The extension reuses the session already on this machine — no API key, no second login.

  1. 01

    Open Extensions

    CmdShiftX · CtrlShiftX

  2. 02

    Search and install

    Paste the name, open Open VSX, or run the command below in a terminal.

  3. 03

    Open the dashboard

    Click Current or Today, or run Cursor Cost: Open Dashboard.

Open VSX
Cursor Cost Tracker
cursor --install-extension lukasz0303.cursor-cost-tracker

Before you install

Does Optimize read the chat?

No. The prompt is built from usage metadata for the last red query. The transcript stays in Cursor. A projected save shows up only after the agent writes .ai/optimize-savings.md.

Pro, or a team plan?

Pro shows included percent for Cursor Models and Other Models. Team, Business, and Enterprise show dollars and one Spend series. The same chart draws used, forecast, and ideal either way.

What do I paste?

Nothing. The extension reuses the Cursor session already on this computer. There is no API key and no second login.

What if Cursor changes the usage API?

The extension is unofficial. If that API moves, the numbers can stall until a release catches up. The token never goes to the panel or the logs.

Support

Fuel the tracker.

If Cursor Cost Tracker helps you catch expensive queries and save tokens, a coffee for the author goes a long way — it keeps late-night status-bar fixes and monthly forecasts brewing.

Buy Me a Coffee

Say thanks with a coffee

One-time tip. Perfect when Optimize or the forecast just paid for itself. Opens in your browser.

Buy me a coffee

Write a message

Comment, feature, or bug

Inside the extension Support tab you can send a note straight to the author — Comment, New feature, Bug report, or Other. Your mail app stays closed.

Open GitHub Issues

Privacy

Your session stays on this machine.

No pasted access token. No proxy. Usage calls go straight to cursor.com using the Cursor session that already exists locally. The panel and logs never show the token.

Optimize builds a prompt from usage metadata only. Projected savings appear when the agent writes .ai/optimize-savings.md in the project — not from a transcript scrape.