Swiss Environment MCP

🇨🇭 Part of the [Swiss Public Data MCP Portfolio](https://github.com/malkreide)

🌿 swiss-environment-mcp

MCP server connecting AI models to Swiss environmental data from BAFU – air quality, hydrology, natural hazards, wildfire danger and open environmental datasets.

🇩🇪 Deutsche Version

---

Overview

swiss-environment-mcp gives AI assistants like Claude direct access to real-time environmental data from Swiss federal authorities – no API keys required. Air quality readings from the national NABEL monitoring network, hydrological gauging stations, natural hazard bulletins, and the full BAFU dataset catalogue are all accessible through a single standardised MCP interface.

The server covers four thematic clusters: air quality (NABEL), hydrology, natural hazards, and the BAFU open data catalogue. Each cluster maps to a group of purpose-built tools that translate raw agency data into clean JSON responses.

Anchor demo query: *"What is the current air quality at the NABEL station Zürich-Kaserne – and does it comply with WHO 2021 guidelines?"* → More use cases by audience →

---

Features

• 🌬️ Air quality monitoring – 16 NABEL stations, NO₂/O₃/PM10/PM2.5/SO₂/CO, Swiss LRV + WHO 2021 limit checks
• 💧 Hydrology – water levels, flow rates, temperatures across Swiss gauging stations
• 🚨 Flood warnings – active alerts filtered by danger level and canton
• 🏔️ Natural hazard bulletin – SLF/BAFU bulletin in DE/FR/IT/EN, region-specific warnings
• 🔥 Wildfire danger – canton- and region-level fire danger index
• ❄️ Snow & avalanches (SLF) – snow depth, new snow per IMIS station; avalanche danger levels (EAWS)
• 🦌 Hunting statistics – cull & game-loss figures per species, canton and year (federal hunting statistics)
• 📦 BAFU open data catalogue – search and retrieve environmental datasets via CKAN
• 🔑 No authentication required – all data sources are publicly accessible
• ☁️ Dual transport – stdio for Claude Desktop, Streamable HTTP/SSE for cloud deployment

---

Prerequisites

• Python 3.11+
• No API keys needed – all endpoints are publicly accessible without authentication

---

Installation

# Clone the repository

git clone https://github.com/malkreide/swiss-environment-mcp.git cd swiss-environment-mcp

Install

pip install -e .

Or with uvx (no permanent installation):

uvx swiss-environment-mcp

Or via pip:

pip install swiss-environment-mcp

---

Quickstart

# Start the server (stdio mode for Claude Desktop)

swiss-environment-mcp

Try it immediately in Claude Desktop:

*"What is the current air quality at NABEL station Zürich-Kaserne?"*

*"Are there any active flood warnings in Switzerland right now?"*

*"What is the wildfire danger level in Canton Valais?"*

---

Configuration

Claude Desktop

Minimal (recommended):

{

"mcpServers": { "swiss-environment": { "command": "uvx", "args": ["swiss-environment-mcp"], "env": {} } } }

Config file locations:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

After saving, restart Claude Desktop completely.

Cloud Deployment (SSE for browser access)

For use via claude.ai in the browser (e.g. on managed workstations without local software):

Render.com (recommended):

1. Push/fork the repository to GitHub
2. On render.com: New Web Service → connect GitHub repo
3. Render detects render.yaml automatically
4. In claude.ai under Settings → MCP Servers, add: https://your-app.onrender.com/sse

Docker:

docker build -t swiss-environment-mcp .

docker run -p 8000:8000 swiss-environment-mcp

💡 *"stdio for the developer laptop, SSE for the browser."*

---

Available Tools

All tools share the stable env_ name prefix — a deliberate namespace choice so that the server's tools are recognisable and unlikely to collide when several MCP servers are mounted together. Tool definitions (name, description, input schema) are pinned via tool-snapshot.json; changes require a CHANGELOG entry (see CONTRIBUTING).

Tool budget (21 tools, 7 clusters) — exhausted. Every tool maps to a distinct user question, not to a REST endpoint — there is no CRUD/endpoint mirroring, and the anchor queries are each answerable in a single call. The count sits above the ≤12 rule of thumb because the server deliberately spans seven environmental domains (air, water, hazards, snow, hunting, catalogue, aircraft noise), each needing a list/detail pair or a domain-specific action. Further consolidation was considered and rejected: the *_stations/*_current pairs (NABEL, hydro, snow) serve genuinely different intents (discovery vs. reading a known station) and collapsing them would overload a single tool's parameters.

This is the ceiling. With the aircraft-noise cluster the server's tool budget is spent: any further data source belongs in a separate *-mcp server, not here. The next addition would instead trigger a review of whether some listings should migrate to MCP resources rather than tools.

🌬️ Air Quality / NABEL (3 tools)

💧 Hydrology (5 tools)

🏔️ Natural Hazards (3 tools)

❄️ Snow & Avalanches / SLF (3 tools)

🦌 Hunting & Wildlife (2 tools)

📊 Environmental Data Catalogue (2 tools)

✈️ Aircraft Noise / BAZL noise cadastre (3 tools)

⚖️ Legal notice — carried in every response of these three tools. The noise

cadastre is an *orientation aid*. Legally binding information on construction

projects is issued by the competent cantonal office or by the FOCA (BAZL).

These tools do not replace a building permit clarification.

Coordinates must be LV95 (EPSG:2056, metres: E ≈ 2'480'000–2'840'000, N ≈ 1'070'000–1'300'000). WGS84 degrees such as 8.54 / 47.37 are rejected fail-fast with a conversion hint — this is the single most common LLM error with Swiss geodata. Convert via swisstopo REFRAME or convert_coordinates in swisstopo-mcp.

The contours are lines, not areas. The cadastre publishes MultiLineString isolines, so identify performs a *proximity* query within a search radius — not a point-in-polygon test. env_noise_aircraft_at therefore returns a bracket («the point lies between the 61 dB and the 62 dB contour») with the highest value as a stated upper bound, never an interpolated point value. The search radius is reported in every response; enlarging it inflates the result (at one Kloten point: 100 m → 61–62 dB, but 500 m → 58–75 dB, because the 75 dB runway contour sits 1.5 km away).

Anchor demo query

**"Is the planned school site inside an aircraft-noise zone with building

restrictions — and at which dB level?"**

Cross-server flow: resolve an address or EGID via swiss-housing-mcp → convert to an LV95 coordinate → env_noise_aircraft_at(east=…, north=…, period="day") → feed the resulting level_db into env_noise_limits_check(level_db=…, sensitivity_level="II", period="day") for the legal classification.

swiss-housing-mcp  →  address / EGID  →  LV95 E/N

↓ env_noise_aircraft_at → 62 dB (upper bound, LBK Zürich, valid from 03.07.2015) ↓ env_noise_limits_check → immission limit ES II (60 dB) exceeded by 2 dB

Example Use Cases

| *"Is the water quality at Strandbad Küsnacht