convez.ai
MCP ServerProdutividadeIntermediário
developer-hasm

Claude Code Dashboard

developer-hasm/claude-code-dashboard

Dashboard local para monitorar uso, custos e gerenciar recursos do Claude Code

Ver no GitHub
InicialPluginScore · de 0 a 100

Painel web local que monitora limites de taxa, gerencia agentes/skills/hooks/servidores MCP e rastreia histórico de sessões no Claude Code. Oferece visualização de custos API por modelo, atividade diária e rankings de uso, com suporte a busca full-text em sessões e detecção de dependências faltantes.

Instalação

$npx convez-ai developer-hasm/claude-code-dashboard

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
1
Saúde
42%
Último commit
há 2 dias
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Monitorar consumo de API e custos por agente/skill em tempo real
  • Gerenciar inventário de agentes, skills, hooks e servidores MCP
  • Rastrear histórico de sessões com visualização de conversas completas
  • Analisar atividade diária e identificar recursos mais utilizados

Veredito

Ferramenta essencial para quem usa Claude Code em produção, oferecendo visibilidade completa sobre consumo de API, custos e gestão de recursos. Implementação técnica sólida com SQLite incremental, suporte a i18n e interface responsiva — pronto para uso via marketplace.

README

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

Claude Code Dashboard

Local web dashboard for Claude Code — monitor rate limits, manage agents/skills/hooks/MCP servers, browse session history, track daily activity and usage rankings.

Status License Next.js React i18n

Demo

Screenshots

Overview

Cost Breakdown by project / agent / skillFull-text session search
Cost BreakdownSession Search

MCP Servers

Features

Overview

  • Rate Limit Monitor — Real-time 5h / 7d window usage from Anthropic OAuth API
  • Cost Estimator — Today's API-equivalent cost by model (Opus / Sonnet / Haiku)
  • Daily Activity Chart — Per-day messages, sessions, tool calls (Activity / Cost toggle)
  • Top 5 Rankings — Most-used agents, skills, and tools (with crown badges)
  • Missing Required Items — Detects skills referenced by agents but not installed

Inventory Management

CategorySource
Plugininstalled_plugins.json + marketplace plugin.json
Agent.claude/agents/*.md (frontmatter parsed)
Skill.claude/skills/*/SKILL.md
Hooksettings.json hooks section
Command.claude/commands/*.md
MCP Serverclaude mcp list CLI output
Sessionsessions/*.json + JSONL transcripts

Each item card shows: name, scope (PROJECT/GLOBAL), description, created/last-used dates, delete button. Sort order: last used DESC → created DESC → name ASC. Delete shows dependency warnings (bidirectional dependents/dependencies).

Session Browser

  • Session cards with first prompt, token counts, cost, context saturation
  • "Messages" click → full conversation modal
  • Scheduled-task badges, start/end times

v1.1 — SQLite Incremental Scanner

  • better-sqlite3 with WAL mode
  • Incremental scan via mtime + file_size comparison — only new lines parsed
  • message_id-based deduplication (UPSERT)
  • Per-model cost estimation stored in turns.cost_usd
  • Schema migration framework

UX

  • Dark mode, EN/KO i18n
  • Responsive layout (768px breakpoint)
  • Vercel-style design (CSS custom properties)

Quick Start

Prerequisites

Install via Claude Code Plugin Marketplace (recommended)

Inside Claude Code, run:

/plugin marketplace add developer-hasm/claude-code-dashboard
/plugin install claude-code-dashboard@developer-hasm

Then use these slash commands inside Claude Code:

CommandDescription
/claude-code-dashboard:openOpen the dashboard in the browser (starts the server if not running). Idempotent. On cold start (server not already up) it also checks for new releases — best-effort, silent on network failure.
/claude-code-dashboard:stopStop the running dashboard server
/claude-code-dashboard:statusShow running status, PID, port, uptime
/claude-code-dashboard:updatePull the latest released version via the marketplace, stop the running server, and print restart instructions

The first :open will run npm install and npm run build in the plugin's install directory (one-time, ~1 minute), then launch the server and open your browser. Subsequent invocations just refocus the existing dashboard.

Updating

On cold start, :open prints a notice when a newer release is available (the exact wording is in commands/open.md):

📦 Update available: 1.0.0 → 1.1.0
   To update, run /claude-code-dashboard:update
   (this dashboard will keep working on the current version until you do)

The check only runs on cold start (when :open actually launches the server) and is best-effort — silent on network failure. When the server is already running, :open skips the version check entirely to keep re-open sub-second.

:update runs claude plugin marketplace update + claude plugin update, stops the current server gracefully (with a forced fallback after 2s), and tells you to restart Claude Code so the new slash commands and install path are picked up.

Install from source

git clone https://github.com/developer-hasm/claude-code-dashboard.git
cd claude-code-dashboard
npm install
npm run build
npm start

The dashboard auto-opens at http://127.0.0.1:19280 (port range 19280–19289).

Usage

The server walks up from the current working directory to find a .claude/ folder. Run npm start from inside (or above) your Claude Code project to see its inventory.

cd /path/to/your/project    # has .claude/ subfolder
npm start --prefix /path/to/claude-code-dashboard

CLI flags:

  • --no-open — don't auto-launch browser

Tech Stack

  • Framework: Next.js 15 (standalone), React 19, TypeScript 5
  • Styling: Tailwind CSS v4 (CSS custom properties)
  • Database: better-sqlite3 (WAL mode)
  • Parsing: gray-matter (YAML frontmatter)
  • Testing: Playwright (25 E2E tests)

Architecture

server.mjs                  → singleton/PID + port allocation + Next standalone fork
src/app/api/*/route.ts      → REST endpoints (inventory, health, token, sessions, ...)
src/lib/scanner.ts          → category scanners (agents, skills, hooks, MCP, etc.)
src/lib/incremental-scanner → SQLite-backed JSONL incremental ingestion
src/lib/cost-estimator      → per-model token-to-USD pricing
src/lib/claude-oauth        → rate-limit fetch via .credentials.json token
src/components/             → React UI (sidebar, ItemCard, modals, charts)

Testing

npm start &           # start server in background
npm run test:e2e      # run Playwright suite

25 E2E tests covering:

  • Smoke (page load, sidebar, API endpoints)
  • Modals (Export, Import)
  • Theme/language toggles
  • Each category page (Plugin, Agent, Skill, Hook, Command, MCP, Session)
  • Item card rendering, dates display
  • Delete dependency warnings

Known Limitations

  • Cold scan: First run with no SQLite cache is slow (mitigated by background scan)
  • MCP last-used: Cloud MCP servers don't log to JSONL, so last-used can't be tracked
  • Cost estimation: API-equivalent pricing — actual Pro/Max plan billing differs

License

MIT — see LICENSE

Ver o original no GitHub

Bom para

Relacionados