🌸 japan-seasons-mcp

Give your AI assistant live Japan travel data — cherry blossom forecasts, autumn leaves, fruit picking, flowers, festivals & more.

1,700+ spots. 17 tools. Live data from Japan Meteorological Corporation.

Live map · Hosted MCP endpoint · npm

Screenshots

The problem

You ask Claude, ChatGPT, or Gemini: "When should I visit Kyoto for cherry blossoms?"

You get "late March to early April" — the same answer every year, based on training data that has no idea what this winter was like.

The actual bloom date shifts by 2–3 weeks depending on temperatures. Tokyo peaked March 29 in 2024, March 22 in 2023, March 28 in 2022 — a 7-day spread in just two years. Your AI is guessing. This MCP server stops the guessing.

The fix

It connects Claude (or any MCP client) to live forecast data from the Japan Meteorological Corporation — the same data that powers SAKURA NAVI and Japan's major cherry blossom forecast sites. Bloom meters update daily at 9 AM JST; spot observations update when JMC reporters file.

You:    "I'm in Japan April 8–14. Where should I see cherry blossoms?"

Claude: Based on today's JMC forecast data:

        Cities in full bloom during your visit:
        • Sendai (Miyagi) — full bloom forecast Apr 10 (avg Apr 12)
        • Morioka (Iwate) — full bloom forecast Apr 13 (avg Apr 16, running early)
        • Akita — full bloom forecast Apr 14

        For Sendai, top spots include Tsutsujigaoka Park (92% full bloom today)
        and Nishi Park. Morioka Castle ruins are a classic.

        One thing to watch: 40% rain chance Apr 10 in Sendai. Rain speeds up
        petal fall, so Apr 11–12 is probably the safer window.

Real bloom percentages. Real park names. Real weather.

Use it with AI assistants

There are two different ways an AI assistant can use Japan in Seasons:

1. AI search / web browsing can find and cite the public pages, text summaries, and JSON APIs. This is the lowest-friction path for people asking "latest sakura dates" in ChatGPT, Perplexity, Google AI Mode, etc.
2. MCP tool use only works after the user or workspace admin connects the MCP server inside an MCP-capable client. Finding this repo in search does not give the assistant permission to run MCP tools.

For AI search, use these crawlable sources:

• Latest sakura text summary: https://seasons.kooexperience.com/sakura-forecast.txt
• Sakura forecast JSON: https://seasons.kooexperience.com/api/sakura/forecast
• Interactive map and forecast pages: https://seasons.kooexperience.com

Remote MCP endpoint

No package install is needed when the client supports remote/streamable HTTP MCP. Add this as the MCP server/app/connector URL:

https://seasons.kooexperience.com/mcp

For ChatGPT Apps/Connectors, use:

• Name: Japan in Seasons
• Description: Use this for live Japan seasonal travel data: cherry blossom and sakura dates, autumn leaves, flowers, festivals, fruit picking, and weather. Best for current or date-specific Japan travel questions.
• Connector URL: https://seasons.kooexperience.com/mcp

Claude Desktop / Claude Code / stdio MCP clients

Add this to the client's MCP config:

{
  "mcpServers": {
    "japan-seasons": {
      "command": "npx",
      "args": ["-y", "japan-seasons-mcp"]
    }
  }
}

Optional connection preferences supported by the hosted endpoint:

• dateStyle — friendly or iso
• temperatureUnit — celsius or fahrenheit
• includeCoordinates — true or false
• mapLanguage — english or japanese

Self-host

PORT=3000 npx -y japan-seasons-mcp --http
# MCP endpoint: http://localhost:3000/mcp

Optional durable cache for the hosted web map:

JAPAN_SEASONS_CACHE_DIR=/data PORT=3000 npx -y japan-seasons-mcp --http

Point JAPAN_SEASONS_CACHE_DIR at a persistent volume, for example a Railway Volume mounted at /data. This stores the warmed national sakura/koyo all-spots JSON across restarts; without it, the app falls back to in-memory cache and still works.

Why ChatGPT may find it but refuse to use it

ChatGPT Search can discover japan-seasons-mcp as a web result, but web discovery is not the same as connecting an MCP server. If the chat only has web search enabled, it should cite the crawlable forecast pages or JSON API. To run tools like sakura_now, sakura_forecast, or sakura_spots, the MCP endpoint must first be added as a ChatGPT app/connector or configured in another MCP client.

What's covered

1,700+ GPS-tagged spots across 12 seasonal categories.

Tools

Best first-call tools

japan_seasonal_answer — answer a broad traveler question

Use this when someone asks a normal question instead of naming a dataset: "How is the sakura forecast?", "What is good in Japan in June?", "Where should I see autumn leaves in late November?", or "What seasonal activities match my dates?" It routes to the right live or curated dataset and returns a ready-to-use recommendation.

sakura_now — current cherry blossom answer

The best first call for broad sakura prompts. It summarizes what is good now, what is coming next, or what matched the user's trip dates, using live JMC forecast/observation data, and includes a short list of specific viewing spots when spot data is available.

koyo_now — current autumn leaves answer

The autumn equivalent of sakura_now: current koyo timing, maple/ginkgo peaks, trip-date matches, and next-step guidance for exact spots.

search / fetch — retrieval-friendly access

Provider-friendly search/fetch tools for ChatGPT-style retrieval and deep-research citation flows. search returns result IDs; fetch returns full text with a canonical URL.

Cherry blossom

sakura_forecast — the big picture

All 48 JMA observation cities in one call: this year's forecast, actual dates when observed, and how each city compares to the historical average. Good starting point before you drill into specific spots.

"What's the cherry blossom situation in Japan right now?"
→ 48 cities by region, bloom status, forecast dates, days vs average

sakura_spots — specific parks and temples

1,012 spots across Japan with current status, bloom percentages, and GPS coordinates. When JMC spot reporters have filed a recent update (within 48 hours), the tool uses that observed status as the primary reading. Otherwise it falls back to the JMC bloom-meter forecast. Stale observations are shown as context, not hidden.

"Cherry blossom spots in Kyoto"
→ 51 spots: Kiyomizu-dera (Full bloom, observed Apr 9), Maruyama Park (91% full-bloom)...

sakura_best_dates — match travel dates to bloom

Give it your start and end dates, get back the cities where full bloom overlaps your window plus a short list of specific viewing spots when spot data is available.

"I'm in Japan April 8–14, where should I go?"
→ Cities with bloom in that window, ranked by timing, plus spot suggestions and map links

kawazu_forecast — early-season deep-pink variety

Kawazu cherry blooms January–February in Izu Peninsula, months before standard sakura opens anywhere.

"Can I see cherry blossoms in February?"
→ 9 Kawazu spots with bloom %, GPS, forecast dates

Autumn leaves

koyo_forecast — maple and ginkgo timing by city

50+ cities with this year's colour-change dates and how they compare to the historical normal. Maple and ginkgo peak at different times; both are included.

"When do autumn leaves peak in Kyoto vs Hokkaido?"
→ City-by-city maple/ginkgo dates, days early or late vs average

koyo_best_dates — same idea as sakura best dates, for autumn

Match your travel window to cities in peak colour.

"I'm in Japan late October, where for autumn leaves?"
→ Cities in peak colour during your dates, maple vs ginkgo timing

koyo_spots — 687 viewing spots by prefecture

Each spot has a peak window (start, peak, end), leaf type, popularity rating, and GPS.

"Top autumn leaves spots in Kyoto"
→ Arashiyama, Eikando, Tofukuji, Rurikoin... with star rating and exact peak dates

Flowers

flowers_spots — 90 curated spots, 8 flower types, Jan through Oct

Filter by type, prefecture, or month. Each spot has an official URL and verified GPS.

Festivals and events

festivals_list — 46 major recurring events with official URLs and attendance figures

"Best fireworks festivals in Japan?"
→ Sumida River (900k), Nagaoka (1.1M), Omagari, PL Osaka, Miyajima...

"Festivals in Kyoto in October?"
→ Jidai Matsuri (Oct 22), Kurama Fire Festival, with booking tips

Filter by type (fireworks / matsuri / winter), month, and prefecture.

Fruit picking

fruit_seasons — full-year calendar for 14 fruits

Which fruits are in season and at peak for any given month, with best regions and notes.

"What fruit can I pick in September in Japan?"
→ Grape at peak (Yamanashi, Nagano), Pear at peak, Peach ending, Apple starting

fruit_farms — 346 farms with GPS and booking links

Pass month= and it auto-filters to farms with something in season. Add region= to narrow further.

"Strawberry farms near Tokyo in April"
→ Farms in the Tokyo/Kanto area with strawberry in season, GPS + Jalan links

Weather

weather_forecast — 3-day JMA forecast for 51 cities

Temperature, rain probability by 6-hour window, and conditions. Worth checking because rain speeds up petal fall.

"Weather in Osaka this weekend?"
→ Min/max temp, rain % per 6-hour window, conditions

Usage

Ask your MCP client for a goal, not a tool name. A few good examples:

"I'm in Japan April 8-14. Where should I go for cherry blossoms?"
"Top autumn leaves spots in Kyoto in late November"
"What flowers are in season in Japan in June?"
"Best fireworks festivals in Japan in August"
"Fruit picking near Tokyo in May"
"Will rain in Osaka this weekend make sakura worse?"

Typical workflow:

1. Ask for timing first with sakura_best_dates, koyo_best_dates, sakura_forecast, or koyo_forecast.
2. Drill into exact parks, temples, farms, or events with sakura_spots, koyo_spots, fruit_farms, flowers_spots, or festivals_list.
3. Check weather_forecast if rain or temperature could change the recommendation.
4. Set optional connection preferences if you want ISO dates, Fahrenheit weather, Japanese map links, or outputs without GPS coordinates.

How it works

flowchart LR
    subgraph live["Live APIs (cached 1–6h)"]
        JMC["Japan Meteorological Corp\nsakura · koyo · kawazu\n1,700+ spots · daily 9AM JST"]
        JMA["Japan Met Agency\nweather · 51 cities\nhourly"]
    end

    subgraph static["Static datasets (loaded at startup)"]
        DATA["flowers.json — 90 spots\nfestivals.json — 46 events\nfruit-farms.json — 346 farms"]
    end

    subgraph server["japan-seasons-mcp"]
        MCP["17 tools\n2 prompts\nstdio + HTTP transport"]
    end

    subgraph clients["MCP clients"]
        C1["Claude Desktop\n/ Claude Code"]
        C2["Cursor\n/ Windsurf"]
        C3["Any MCP\nclient"]
    end

    JMC -->|live fetch| MCP
    JMA -->|live fetch| MCP
    DATA -->|in-memory| MCP
    MCP -->|MCP protocol| C1
    MCP -->|MCP protocol| C2
    MCP -->|MCP protocol| C3

Static datasets load at startup and are served from memory with no disk I/O per request. Live JMC data is cached server-side (1–6h TTL). The all-spots payload is pre-gzipped at startup so repeat serving is essentially free.

Bloom scale reference

JMC publishes two separate data products for sakura spots. Both are used:

Spot observations — reported by JMC partners and spot managers, used as primary status when updated within 48 hours:

State 0  Pre-bloom (buds visible)
State 1  First bloom — 開花 (a few flowers open)
State 2  30% bloom — 三分咲き (sanbu-zaki)
State 3  70% bloom — 七分咲き (nanabu-zaki)
State 4  Full bloom — 満開 (mankai)
State 5  Petals starting to fall — 散り始め
State 6  Green leaves — 葉桜 (hazakura, bloom season over)

Bloom-meter forecast (jr_data) — mathematical model used as fallback when no fresh observation exists:

BLOOM RATE — progress toward first bloom (開花)
─────────────────────────────────────────────────
 0%      60%       85%      100%
 │  Bud   │ Swelling │ Opening │ <- First bloom!
 花芽〜つぼみ  膨らみ始め   開き始め    開花

FULL BLOOM RATE — progress toward mankai / 満開
─────────────────────────────────────────────────
 0%   20%   40%   70%   90%  100%
 │Open│ 30% │ 50% │ 70% │Full │ <- Mankai!
 開花  三分咲き 五分咲き 七分咲き 満開

The forecast model stays frozen at full-bloom=100% after peak and cannot detect petal fall or hazakura on its own. Spot observations (states 5–6) are the only way to confirm post-peak status for a specific park.

Peak viewing is typically full bloom ± 3 days. Rain accelerates petal fall.

Web app

seasons.kooexperience.com is the interactive companion to this MCP server. It shows all the same data on a map — 1,012 sakura spots with lifecycle colours (orange bud, pink bloom, green ended), 687 koyo spots, 346 fruit farms grouped by location, and 90 flower spots. There are also focused SEO/citation pages for cherry blossom forecasts, autumn leaves forecasts, and the MCP server. The map includes a "Plan My Trip" mode where you pick cities and see every seasonal activity near each one ranked by distance, and a "Near Me" button that finds spots within 30km of your GPS location.

Development

git clone https://github.com/haomingkoo/japan-seasons-mcp.git
cd japan-seasons-mcp
npm install
npm run build
npm start            # stdio MCP mode
npm run start:http   # HTTP mode, MCP at http://localhost:3000/mcp

TypeScript. No external database. No auth required.

Data sources

Contributing

PRs welcome, especially for flower spots, festival entries, and farm corrections. See CONTRIBUTING.md.

Formerly

Previously published as japan-sakura-koyo-mcp (deprecated). Use this package instead:

npx -y japan-seasons-mcp

License

MIT · Built by Haoming Koo