Climate Impacts MCP Server

An open-source MCP (Model Context Protocol) server that gives LLMs access to climate change impact data. It wraps the Climate Impact Explorer API so that AI assistants like Claude can answer questions about temperature change, drought, heatwaves, agriculture, and more — by country, emission scenario, and time horizon.

Data Attribution

All climate impact data is provided by the Climate Impact Explorer, developed by Climate Analytics and based on climate impact models developed by IIASA (International Institute for Applied Systems Analysis).

This MCP server is a translation layer — it does not store or modify the underlying data.

Features

• Single-call country overview — ask "what are the climate impacts in Costa Rica?" and get a multi-variable summary in one tool call
• Timeseries projections — year-by-year data from 2015 to 2100 with uncertainty ranges
• Scenario comparison — side-by-side comparison of emission pathways (current policies, 1.5C, NDCs, etc.)
• Warming level snapshots — impacts indexed by global warming level (1.5/2.0/2.5/3.0C)
• Spatial map data — per-cell gridded data with country boundary (TopoJSON) for map rendering
• Smart validation — fuzzy-matching on country names, variable IDs, and scenarios with helpful suggestions

Tools

Tool	Description
get_country_overview	Comprehensive multi-variable impact summary for a country (accepts names like "Costa Rica")
lookup_country	Fuzzy-match country name to ISO code
list_climate_variables	List available climate variables by category
list_scenarios	List emission scenarios
get_climate_projections	Timeseries projections for a country/variable/scenario
compare_scenarios	Compare multiple scenarios side-by-side at key time horizons
get_warming_level_snapshot	View impacts by warming level (1.5/2.0/2.5/3.0C)
get_spatial_data	Gridded spatial data + country boundary for map rendering

Quick Start

Option 1: Use the hosted server (no install required)

Available on Smithery — install with one click directly from the Smithery UI.

Alternatively, add it manually to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "climate-impacts": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://climate-impacts-mcp.vlemaire.com/mcp"]
    }
  }
}

Restart Claude Desktop. The climate tools will appear automatically. No API keys or accounts needed.

mcp-remote is a small bridge that connects Claude Desktop to remote MCP servers. It is installed automatically by npx on first run (requires Node.js).

Option 2: Run locally

1. Install dependencies:

   poetry install
   

2. Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

   {
     "mcpServers": {
       "climate-impacts": {
         "command": "/path/to/your/virtualenv/bin/python",
         "args": ["-m", "climate_impacts_mcp"]
       }
     }
   }
   

   Replace the command path with the output of poetry env info -e.

3. Restart Claude Desktop. The climate tools will appear automatically.

Option 3: Self-host (Streamable HTTP transport)

PORT=8080 poetry run climate-impacts-mcp

The server runs in Streamable HTTP mode when PORT is set, suitable for Cloud Run or any remote deployment. The default remote MCP endpoint is /mcp.

Architecture

src/climate_impacts_mcp/
├── server.py          # FastMCP entry point, lifespan (httpx client + caches)
├── client.py          # Async httpx wrapper for CIE API v2
├── models.py          # Pydantic models for API responses
├── formatting.py      # Markdown formatting for LLM-friendly output
├── boundaries.py      # Country boundary extraction from world-atlas TopoJSON
└── tools/
    ├── metadata.py    # Discovery tools (cached): lookup_country, list_climate_variables, list_scenarios
    ├── timeseries.py  # Projection tools: get_climate_projections, compare_scenarios, get_warming_level_snapshot
    ├── geodata.py     # Spatial tool: get_spatial_data (grid + boundary)
    ├── overview.py    # High-level: get_country_overview (multi-variable parallel fetch)
    └── validation.py  # Input validation with fuzzy-matching

On startup, the server pre-fetches and caches:

• CIE API metadata (countries, variables, scenarios)
• World-atlas TopoJSON (country boundaries for map rendering)

Deploy to GCP Cloud Run

Using Cloud Build

gcloud builds submit --config cloudbuild.yaml

If you want the service to be publicly invokable, run this once after deployment:

gcloud run services update climate-impacts-mcp \
  --region us-central1 \
  --no-invoker-iam-check

Manual deploy

gcloud run deploy climate-impacts-mcp \
  --source . \
  --region us-central1 \
  --port 8080

To make the deployed service public:

gcloud run services update climate-impacts-mcp \
  --region us-central1 \
  --no-invoker-iam-check

Docker

docker build -t climate-impacts-mcp .
docker run -p 8080:8080 climate-impacts-mcp

Development

# Install all dependencies (including dev)
poetry install

# Run tests
poetry run pytest

# Lint
poetry run ruff check src/ tests/

# Interactive MCP inspector
poetry run mcp dev src/climate_impacts_mcp/server.py

Configuration

Environment Variable	Default	Description
PORT	(unset)	If set, runs in SSE mode on this port. If unset, runs in stdio mode.

License

MIT — see LICENSE.

Acknowledgments

• Climate Analytics — Climate Impact Explorer API and data
• IIASA — Climate impact models underlying the data
• Natural Earth / world-atlas — Country boundary data for map rendering