convez.ai
PluginDevOps & InfraestruturaIntermediário
nikhilnt1234

TokenBurnRate

nikhilnt1234/TokenBurnRate

Rastreie, visualize e otimize gastos com tokens de IA em tempo real

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP server + CLI que rastreia cada chamada de API (Claude, GPT, Gemini) localmente, exibe um dashboard de custos no terminal e fornece otimizações específicas para reduzir gastos com tokens. Inclui engine de hints determinísticos que identifica automaticamente modelos caros, subutilização de cache e padrões de consumo ineficientes.

Instalação

$npx convez-ai nikhilnt1234/TokenBurnRate

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Monitorar custos de API de múltiplos modelos de IA em um único dashboard
  • Identificar oportunidades de economia automáticas com base em padrões de uso
  • Integrar rastreamento de tokens em workflows do Claude Desktop

Veredito

Ferramenta prática e bem-concebida para quem usa múltiplas APIs de IA. O engine de hints (8 regras determinísticas) é direto e útil, sem overhead de LLM. A integração MCP + CLI oferece flexibilidade real, mas requer Node.js ≥22 e setup manual em algumas plataformas.

README

A tradução para o português deste README ainda está na fila.

TokenBurnRate

See where your AI tokens go — and how to spend less of them.

npm npm downloads License: MIT Node ≥22


TokenBurnRate is an MCP server + CLI that logs every Claude / GPT / Gemini API call locally, shows you a cost dashboard in your terminal, and tells you exactly how to reduce that cost.

  ⬡ token-tracker  LAST 7 DAYS
────────────────────────────────────────────────────────────────────────

  Overview
  Total cost          $18.7421   (~$80.59/month est.)
  API calls           347
  Input tokens        4.82M
  Cache reads         620K  (11.4% hit rate)

  Daily Cost
  Mon Jun 02  ████████████░░░░░░░░░░░░   $2.14
  Tue Jun 03  ████████████████████░░░░   $3.82
  Thu Jun 05  ████████████████████████   $4.51

  💡 Optimization Hints               saves $31.20/mo
  ● CRIT  Prompt cache barely used
          Cache hit rate: 11.4% — target is 30–60%
          Action: Move static content to top of messages
          Est. saving: $12.40/month

  ● HIGH  Expensive model doing test generation
          104 test-gen calls on Sonnet costing $2.25/week
          Action: Route to claude-haiku-4-5, saves 80%
          Est. saving: $7.80/month

Install

npm install -g token-tracker-mcp

Or run without installing:

npx token-tracker-mcp report

Add to Claude Desktop

node scripts/setup.js

Or add manually to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "token-tracker": {
      "command": "token-tracker",
      "args": ["serve"]
    }
  }
}

Restart Claude Desktop. Done.


CLI Commands

CommandDescription
token-tracker reportFull 7-day dashboard
token-tracker report --period monthMonthly report
token-tracker todayToday only
token-tracker hintsOptimization hints ranked by $ saving
token-tracker hint <id>Deep-dive on one hint
token-tracker statusOne-line: cost · cache % · top hint
token-tracker budgetBudget gauges
token-tracker modelsPricing table for all models
token-tracker export > out.csvRaw CSV export

MCP Tools (use inside Claude)

ToolDescription
log_usageLog an API call — auto-calculates cost
get_summarySummary for today / week / month / all
get_hintsRanked optimization hints with $ savings
get_hint_detailDeep-dive on a specific hint
set_budgetSet a daily / weekly / monthly spend limit
list_sessionsSessions ranked by cost
list_modelsPricing table
export_csvCSV dump

Optimization Hints Engine

8 deterministic rules — no LLM calls, runs instantly on your local data:

HintTriggers when
cache-utilizationCache hit rate < 30%
model-swap-testingTest gen running on Sonnet / Opus
model-swap-debugDebugging on Opus
verbose-outputsOutput / input ratio > 0.35
session-spikeAny session costs 3× your average
context-bloatAvg tokens / call > 8K
retry-loopsSessions with 30+ high-token calls
single-model-dependency100% traffic on one expensive model

Each hint includes severity · evidence · recommended action · estimated monthly saving.


Privacy

All data stored at ~/.token-tracker/usage.db (SQLite). Nothing leaves your machine. No telemetry, no account required.


Development

git clone https://github.com/nikhilnt1234/TokenBurnRate.git
cd TokenBurnRate
npm install --ignore-scripts
npx tsup
npm test

Roadmap

  • Team / Supabase backend (multi-user shared dashboard)
  • Weekly email digest
  • Slack / webhook alerts
  • macOS menubar app

License

MIT © 2026 Nikhil T

Ver o original no GitHub

Bom para

Relacionados