AI agents and the MCP server
Everything the Criora map and Atlas show can be read by an AI assistant or by an agent you build: the seven-day forecast and its risk for any point, the long-term climate profile of any point, hazard events near a point, and the risk profile of every country. Criora offers it through an MCP server.
No account or key is needed. The server is read-only.
For AI assistants: the MCP server
Criora runs a remote Model Context Protocol server at:
https://criora.com/mcpAdd it as a custom connector in any assistant that supports remote MCP servers (Claude, ChatGPT, Cursor, VS Code and others). No authentication is required. For example, in an MCP configuration file:
{
"mcpServers": {
"criora": { "url": "https://criora.com/mcp" }
}
}In VS Code or Cursor, add it in one click: VS Code, Cursor.
To give an agent you build the same tools, see Agent frameworks: LangChain, LlamaIndex, the OpenAI Agents SDK and n8n.
The server offers the tools below. The place tools take latitude and longitude, or a place name that they resolve to the best match and report back.
| Tool | What it answers |
|---|---|
assess_place | Up to ten places, each on the days asked, in one answer: forecast risk by day, the month’s 1991-2020 normal past the forecast, hazards nearby, climate profile, country |
get_forecast | Seven-day weather and weather risk, by local day or step by step |
rank_by_forecast | The cities or countries where the weather of the next 1 to 7 days is most dangerous, optionally for one hazard |
get_area_risk | The weather of the next 1 to 7 days over a whole country or region, day by day, and on request its long-term climate layers |
get_climate_profile | Long-term exposure: floods, sea level rise, heat, drought, water stress and more; warming and rainfall change as the models’ median with their 10th to 90th percentile |
get_climate_table | Long-term layers at 2 to 50 places in one call, as a table: a row per place, a column per layer and band; the headline exposures unless categories or layers are given; each row with a home’s damage band |
get_climate_twin | The large city whose climate today a city will have in 2041-2060, under two scenarios |
get_place_history | What has hit a place in past decades: earthquakes, cyclones, floods, tsunamis and eruptions, by decade, the largest with their date, distance, deaths and damage, and from which year each count is complete |
estimate_financial_impact | Indicative flood and cyclone damage to a building or asset: its expected annual damage and the damage of the 50 to 500-year events, as a share of its value and in money, low to high, today or in 2050 and 2080 |
get_hazards_near | Wildfires, earthquakes, floods, cyclones, volcanoes, droughts and air quality near a place |
list_hazards | Disasters and official warnings in force in a country, a region or the whole world |
get_country_profile | A country’s risk scores and, on request, its indicators furthest from the world median, or all of them, with their benchmarks |
get_enso_outlook | NOAA’s outlook for El Niño and La Niña over the seasons ahead and, for a country, what each has brought it in those seasons since 1950 |
compare_countries | Two to ten countries side by side, with their ranks |
rank_countries | Countries most or least at risk, overall, on one dimension or on one hazard or other indicator, optionally within a region or income group |
find_place | Coordinates for a place name, when a name is ambiguous |
search | Countries, places, Criora’s methodology and data sources pages and its catalogue of layers and indicators matching a query, which may name a place and a subject together, as ids for fetch and pages to cite |
fetch | One country, place, page or catalogue as a markdown document with the Criora page to cite, for research agents such as ChatGPT deep research |
show_risk_map | An interactive map in the conversation, of a place, several places side by side, one event such as a storm, or the world: this week’s risk, or the rain, temperature or wind forecast, under the hazards nearby |
In assistants that display MCP Apps, such as Claude, show_risk_map draws the map inside the conversation: the reader can pan and zoom, open it full screen, play the week hour by hour and switch between the forecast risk, rain, temperature and wind. It opens on a place, on several places with each one’s week side by side, on one event such as a storm with its forecast cone and track, or, given none of these, on the world, with the most critical events and the countries where most people are at risk this week. Assistants that do not display apps get the same answer as text, with a link to the place on Criora’s map.
The server also offers five prompts, ready-made questions an assistant can start from: assess_trip for how risky a trip is on its dates, compare_places_to_live for the long-term exposure of several places side by side, country_briefing for one country against its region and the world, site_due_diligence for the climate and disaster risk of a property or site, and esg_country_screen for screening countries on climate and disaster risk for an ESG, investment or supply-chain decision.
Reading the forecast risk
Each forecast step carries a risk class from 0 (Very low) to 4 (Critical), the hazard that sets it, and the physical reading behind it: the UTCI felt temperature for heat and cold, the gust speed for wind, the precipitation rate for rain and snow. Rain, snow and wind take precedence over temperature, which is the same rule the map uses. Each day’s summary names its headline risk and lists the highest class every hazard reaches that day.
Reading an event’s severity
Every event carries severity_rank, Criora Severity from 0 to 100, which compares events across hazards: a flood with an earthquake or a fire. It weighs the event’s official level on one common scale, the people living inside its footprint (people, from GHS-POP 2025) and its reading against the others of its type. It was fitted on real past situations by what happened afterwards, and orders 78.5% of their pairs as their outcomes did, against 30% for the order Criora used before.
The official level is always returned beside the rank, never replaced by it, and footprint names how the area people are counted in was drawn. In a detailed answer, building_value is the replacement cost, in euros, of the buildings and their contents inside the same footprint, homes apart from the rest: indicative, not market value. get_hazards_near and list_hazards list the highest rank first. get_hazards_near, and list_hazards for a country, also list in coverage every source that reports there, its hazards and whether it is reporting now or has been silent since its last read; warnings: false means no authority’s warnings are reaching Criora there now. volcano_levels: false means no volcano observatory’s alert levels reach Criora there: a volcano is known only by its eruptions as reported. What the rank weighs and how far it can be relied on: Criora Severity.
Given days, get_hazards_near and list_hazards list every event active in those days, those that have ended included, each marked current: true while it goes on, false once it is over.
What a financial estimate rests on and how far it can be relied on: Financial impact estimates.
Pages as markdown
Every public page of the site, a country or a city of the Atlas, a guide, an event, is also served as markdown: the text a reader sees, without the layout. Ask for the page’s own address with Accept: text/markdown, or add .md to it, as in https://criora.com/atlas/PRT.md. The text opens with the page’s title, description, address and language. The map’s pages answer with llms.txt, the site’s description for agents.
Limits
Place names are searched in Criora’s own gazetteer of GeoNames places (CC BY 4.0): cities, towns, villages, regions, islands, volcanoes and lakes, with their names in every language, but not street addresses. When a name is ambiguous, add the region or country after a comma: Portland, Maine.
Forecast answers change when a new model run is published, twice a day. Climate layers change when a dataset is rebuilt. Fields may be added to answers; existing fields keep their meaning.
Sources and attribution
Every answer names its source. When you publish or redistribute data from Criora, credit Criora and follow the licence of the source it names: the forecast comes from Environment and Climate Change Canada under its Data Server End-use Licence, days 8 to 15 from the ECMWF ensemble under CC BY 4.0 with the credit “Based on data and products of the European Centre for Medium-Range Weather Forecasts (ECMWF)”, place names from GeoNames under CC BY 4.0, street addresses and reverse lookups from OpenStreetMap under the ODbL, and each climate layer lists its own provider and licence. See Data Sources for the full list. Use of the MCP server and of Criora’s data is governed by the Terms of Service.
Criora publishes data, not advice. Its figures are not a substitute for professional financial, insurance or safety advice.
