convez.ai
AgenteProdutividadeIntermediário
fllantada

Claude Travel Planner

fllantada/claude-travel-planner

Planejador de viagens com dados reais, sem alucinações de preços

Ver no GitHub
InicialPluginScore · de 0 a 100

Agent de IA que planeja viagens completas dentro do Claude Code. Pesquisa destinos, busca voos e hotéis reais via SerpApi, gera itinerários dia a dia com justificativas para cada decisão e exporta tudo em PDF profissional pronto para enviar.

Instalação

$npx convez-ai fllantada/claude-travel-planner

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

Casos de uso

  • Gerar roteiros de viagem personalizados com orçamento justificado
  • Pesquisar voos e hotéis em tempo real e consolidar em um plano
  • Criar PDFs de viagem profissionais para enviar a clientes ou amigos
  • Planejar multi-destinos com rotas otimizadas e recomendações baseadas em dados

Veredito

Agent robusto que resolve um problema real: planejar viagens sem alucinações de preços. O design modular com skills separadas e orquestração via Claude Code é elegante, mas exige setup local de ferramentas (pandoc, weasyprint) e chaves de API. Ideal para agências ou profissionais que querem automatizar cotações de viagem.

README

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

🌍 claude-travel-planner

An AI travel agent that lives inside Claude Code. Ask for a trip, and it researches the destination, pulls real flights and hotels, writes a day-by-day plan — with restaurants, activities, a budget, and a reasoned justification for every decision — and renders it as a beautiful PDF it can email to you.

No bespoke agent framework, no servers. Just a handful of declarative skills + two tiny Node scripts, orchestrated by Claude Code.

📄 See it in action: examples/ — two real plans (a 15‑night Iceland Ring Road honeymoon, and a budget‑capped Iceland + Copenhagen version). Same system, different brief.


Why this exists

Most "AI trip planners" hallucinate prices and invent hotels. This one is built around three rules:

  1. Real data for anything with a price — flights & hotels come from the SerpApi (Google Flights / Google Hotels) engines, never made up.
  2. Justify every decision — each choice (destination, route, vehicle, hotel, activity) ships with a "why," grounded in a real fact. Never assume; if a number isn't known, it's marked "to confirm."
  3. Static stuff is researched, not integrated — climate, best season, what to see, itineraries and restaurants come from web research (fanned out across parallel sub‑agents), not from brittle API integrations.

The output looks like something a real travel agency would charge for.


How it works

            ┌─────────────────────────  Claude Code  ─────────────────────────┐
            │                                                                  │
  you ──▶   │  skill: viajes          skill: resumen-viaje      skill: sendgrid│
  "plan a   │  ├ research playbook     ├ quality bar             └ email PDF    │
   trip"    │  │ (parallel agents)     ├ plantilla-plan.md          (optional)  │
            │  └ buscar-vuelos.js      ├ componentes.md                         │
            │    buscar-hoteles.js     └ estilo-viaje.css ──▶ pandoc+WeasyPrint │
            │     (SerpApi, real $)                              ──▶  PDF        │
            └──────────────────────────────────────────────────────────────────┘
  • viajes — the research + real‑data skill. Runs the parallel‑agent research playbook (5 dimensions: destination/climate/season, route, activities, logistics/budget, food) and calls the SerpApi scripts for live flight/hotel prices.
  • resumen-viaje — turns it all into the PDF: a reusable template (plantilla-plan.md), a component cheat‑sheet (componentes.md), and a theme (estilo-viaje.css) for a cover page, day cards, chips and callout boxes.
  • sendgrid — optionally emails the PDF.

Everything is generic: any destination, any dates. The "key experience" of a trip (auroras, whales, a festival, diving…) just swaps into the same template.


Quickstart

Requirements: Claude Code, Node 18+, and pandoc + weasyprint for the PDF.

# 1. Clone
git clone https://github.com/fllantada/claude-travel-planner.git
cd claude-travel-planner

# 2. Install the PDF toolchain (macOS example)
brew install pandoc
pip3 install --break-system-packages weasyprint

# 3. Add your keys (see "API keys" below)
cp herramientas/.env.example herramientas/.env
#   …edit herramientas/.env and paste your real keys…

# 4. Open the folder in Claude Code and just ask:
#    "Armá un plan de 10 días por Japón en abril 2027, salida desde Madrid,
#     presupuesto 5000€ los dos, y mandámelo en PDF."

Claude Code will pick up the skills automatically (they live in .claude/skills/), do the research, search real flights/hotels, write the plan, render the PDF, and (if you set up SendGrid) email it.


🔑 API keys — where to get them

All keys go in herramientas/.env (copied from .env.example). This file is gitignored — never commit it.

KeyWhat it's forWhere to get itCost
SERPAPI_KEYReal flight & hotel prices (Google Flights/Hotels)serpapi.com/manage-api-keyFree — 250 searches/month, no credit card
SENDGRID_API_KEY (optional)Emailing the PDFSendGrid → Settings → API KeysFree tier available
SENDGRID_FROM (optional)The verified "from" addressSender verification—
EMAIL_TO / EMAIL_TO2 (optional)Who receives the planyour own email(s)—

The email step is optional — the PDF is the real deliverable. Without SendGrid you still get the full plan on disk.

The scripts read the key themselves from .env; you never have to paste it into a prompt, and Claude never needs to see its value.


Project structure

claude-travel-planner/
├── .claude/skills/
│   ├── viajes/             # research playbook + SerpApi search (referencia-serpapi.md)
│   ├── resumen-viaje/      # PDF builder: SKILL + plantilla-plan.md + componentes.md + estilo-viaje.css
│   └── sendgrid/           # email the PDF (optional)
├── herramientas/
│   ├── buscar-vuelos.js    # flights  → SerpApi google_flights
│   ├── buscar-hoteles.js   # hotels   → SerpApi google_hotels
│   ├── lib/serpapi.js      # shared helper (reads SERPAPI_KEY from .env)
│   └── .env.example
├── docs/                   # API research notes + architecture
├── examples/               # real sample outputs (PDFs)
└── CLAUDE.md               # how the workspace works (read by Claude Code)

Each trip you plan lives in its own folder at the repo root (e.g. iceland-2026/) with investigacion/, itinerario/, hallazgos/ and the final plan-<trip>.md / .pdf.


Cost & limits

  • SerpApi free tier = 250 searches/month, shared. Each flight/hotel query is one search, so the skills are deliberately economical (static research uses web search, not the quota).
  • Claude Code usage is on your own plan. The example plans were built as a proof‑of‑concept with spare weekly tokens.
  • PDF rendering (pandoc + WeasyPrint) is free and local.

Notes

  • The skills and prompts are written in Spanish (the author's language) — they translate trivially; the structure is language‑agnostic.
  • The example PDFs are real outputs and contain sample personal data; swap in your own trip.
  • This generates a plan to decide and book — it never books or pays for anything.

Security

.env and *.local are gitignored. No keys are committed. The flight/hotel scripts read SERPAPI_KEY from .env at runtime. If you fork this, double‑check your own .gitignore before pushing.

License

MIT — use it, fork it, build your own travel agent.


Built with Claude Code. PRs and ideas welcome.

Ver o original no GitHub

Bom para

Relacionados