Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    TokenOps: Claude Usage, Limits and Savings

    1

    Know where your Claude Code tokens go, and stop paying for a bigger model than the job needs.

    Free

    Security scanned
    TokenOps: Claude Usage, Limits and Savings

    TokenOps: Claude Usage, Limits and Savings

    Example session with this skill installed

    How much have I spent on Claude Code this week and where can I save money?

    • Read your context and instructions
    • Compiled the tokenops

    Summary: Over the last 7 days you spent an estimated

    $36.24 (about $6.04 a day), slightly below the previous week ($37.09). About

    $8.03 (22%) of that could have been saved: 231 requests were short or routine enough for a cheaper model. The biggest gain is moving simple read, search and run turns from Sonnet to Haiku (about $4.74), then using Sonnet instead of Opus for moderate edits (about $3.30).

    TokenOps report

    Last 7 days · 554 requests · 5 sessions · 34 log files. Costs are estimates at API list prices (USD), not a subscription bill.

    Summary

    MetricValue
    Tokens40.25M (92% served from cache)
    Estimated cost$36.24 ($6.04/day avg)
    Output tokens755.0K
    Active days34 of 40 · streak 4 · longest 14

    Limits

    • Daily: $6.85 used (no limit set; pass --daily N) · resets in 3h 42m (Sat 7 Nov, 00:00)
    • Weekly: $29.20 used (no limit set; pass --weekly N) · resets in 2d 3h 42m (Mon 9 Nov, 00:00)

    Compared with the previous 7 days

    • Cost: $36.24 vs $37.09 before (▼ 2%)
    • Tokens: 40.25M vs 42.38M before (▼ 5%)
    • Requests: 554 vs 587 before (▼ 6%)
    • Cache hit: 92% vs 92% before (▼ 0.3 pp)
    • Opus/Fable share: 30% vs 28% before (▲ 1.8 pp)

    Models

    ModelRequestsTokensCostShare
    claude-sonnet-5-539928.86M$24.1467%
    claude-opus-5-5916.69M$10.7130%
    claude-haiku-5-5644.70M$1.394%

    Top projects

    ProjectRequestsCost
    project-1250$15.87
    project-2206$13.57
    project-398$6.80

    Tools

    ToolRequestsCost
    Read256$15.34
    Edit160$11.47
    (no tool)138$9.42

    A request that used several tools counts toward each of them.

    Optimise

    Potential savings $8.03 (22% of $36.24): 231 requests (42%) could have used a cheaper model.

    UsedCould useRequestsSaves
    sonnethaiku162$4.74
    opussonnet45$1.81
    opushaiku24$1.49

    Biggest opportunities by project: project-1 ($3.60), project-2 ($2.90), project-3 ($1.53).

    Recommendations

    • 30% of spend (91 requests) went to Opus/Fable-class models. Make a mid-tier model your default and step up only for architecture or hard debugging.
    • Many requests were short tool calls (read, search, run). Delegate exploration and file reading to a Haiku subagent.
    • Moderate edits and answers (under 3.5K output tokens) are within Sonnet capability. Plan with the big model, implement with Sonnet.

    Last 14 days

    ▇▁█▇▇▇ 2026-11-01 to 2026-11-06 (peak $8.83/day)

    Savings are heuristic: each request is scored by output size, extended thinking and tools used. Try the cheaper tier on routine work first.

    Next steps: (1) Make Sonnet your default model and send file reading and searching to a Haiku subagent: that is the biggest saving. (2) Tell me a daily and weekly budget and I will track it, warn you when you are close and show when it resets. (3) I can also make a shareable one-page HTML report or a CSV of the daily costs.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Know where your Claude Code tokens go, and stop paying for a bigger model than the job needs.

    TokenOps turns the session logs Claude Code already saves on your computer into a clear report your agent can read out and act on.

    WHAT YOU GET

    • Usage and cost. Tokens, requests, sessions, and an estimated cost for any window (last 7, 30 days, or all time), with cache hit rate.

    • Model mix. Spend and share per model (Haiku, Sonnet, Opus, and others) and per project.

    • Daily and weekly limits. Set your own budget; see how much is used, the time left until reset, the exact reset time, and a pace projection with an On track, Pace exceeds limit, or Limit reached status.

    • Limit alerts. An Alerts section appears at the top of the report when a limit is 80% or more used, when it is reached, or when your current pace will hit it before the reset ("at this pace you will reach the daily limit around 22:15"). A one-line check mode returns exit code 0, 1, or 2, and a statusline mode prints a short line you can wire into your own Claude Code status line or hook. A short guide with copy-paste examples is included. The skill itself cannot send notifications or install hooks.

    • Compare with the previous period. Cost, tokens, requests, cache hit rate, and the share of spend on the biggest models, compared to the period just before it, with a warning when your logs only partly cover that earlier period.

    • Tool breakdown. Which tools (Read, Edit, Bash, and others) your requests used and what they cost.

    • CSV export. Daily, model, project, and tool tables as CSV, with the currency in the cost column. Redirect the output to a file with your shell.

    • Optimise. Finds the requests where a cheaper model would plausibly have done the job (short read, search and run turns for Haiku; moderate edits for Sonnet), and totals the potential savings overall and per project, with plain-language recommendations.

    • Activity streaks. Active days, current streak, and longest streak.

    • One-page HTML report. A self-contained, offline page with charts you can open in any browser, screenshot, or print.

    • Currency. USD and INR, whichever works best for you!

    HOW IT WORKS

    Your agent runs a small Node.js script that reads the .jsonl session logs in ~/.claude/projects, removes streamed duplicates, prices each request with built-in (overridable) list prices, and prints a Markdown report. A second script turns the same data into the HTML page, and a third prints CSV, the check line, and the status-line text.

    PRIVATE BY DESIGN

    Read-only. No network calls, no accounts, no uploads. Only model names, token counts, tool names, and timestamps are used; message text is never read. A --hide-projects flag masks project names before you share a report.

    WHERE IT STOPS

    Costs are estimates at API list prices, not your subscription bill. The limits are budget targets you set, because Claude's own plan limits are not stored in the logs. Savings are a heuristic that cannot judge quality, so try the cheaper model on routine work first. Claude Code only: not the claude.ai chat app or the API console.

    Unofficial skill, not affiliated with Anthropic. Claude and Claude Code are trademarks of Anthropic, PBC.

    Requires Node.js 18+. Only supports local Claude Code logs, not claude.ai or direct API usage. Estimates costs based on list prices, not live billing data.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 10 days ago

    • Free with an Agensi account

    Requires

    Claude Code
    NodeJS

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Listed10 days ago

    What's inside

    Frequently Asked Questions

    No, TokenOps is designed specifically for the Claude desktop version. It analyzes local session logs located in your system's Claude projects directory to provide detailed cost and token insights that aren't available through the standard interface.

    No, your data remains completely private. The tool runs as a local Node.js script that parses your local files and generates reports on your machine; it never transmits your session logs or usage data to any external servers.

    The tool applies current Anthropic list prices to your token counts for models like Sonnet and Haiku. While it provides a highly accurate estimate of your spend, it may not account for specific enterprise discounts or regional taxes on your final billing statement.

    You will receive the complete Node.js source files required to run the analysis, a set of command-line instructions, and the templates used to generate the HTML and CSV dashboards.

    Setup is simple: ensure you have Node.js 18+ installed, then point the script to your Claude projects directory. Since it uses standard library modules, you won't need to manage complex external dependencies or API keys for the tool itself.

    TokenOps includes an optimization engine that scans your history for high-cost tasks. It identifies specific tool calls or long conversations that could have been handled by a cheaper model like Haiku without sacrificing performance, helping you adjust your future workflow.

    New in LLM Ops