open-meteo-mcp-server

Public Hosted Server: https://open-meteo.caseyjhand.com/mcp

Tools

Eleven tools covering geocoding, weather forecasts, historical climate, probabilistic ensemble forecasts, marine conditions, air quality, terrain elevation, river discharge, CMIP6 climate projections, and SQL analytics over large datasets:

openmeteo_search_locations

Resolve a free-text place name to ranked coordinate matches. Required first step for name-based queries — all weather tools accept latitude/longitude, not place names.

• Returns name, country, admin1/admin2, latitude, longitude, elevation, IANA timezone, population, and GeoNames feature code

• Search by a bare place name — a city, region, or landmark ("Baoding", not "Baoding Hebei"; "Paris", not "Paris, France"); a compound "City Region" or "City, Country" string matches nothing

• Disambiguate same-named places (e.g., "Springfield") with the optional country filter (ISO 3166-1 alpha-2, e.g. US) or by raising count (default 5, up to 10) and reading the admin1/country fields on each result — those are output fields for choosing among matches, not search inputs

• Pass the timezone from an openmeteo_search_locations result directly to weather tools as the timezone parameter

• Fails with a no_results error (not an empty array) when nothing matches — retry the bare place name without qualifiers, or for a physical feature/landmark search the nearest populated place instead

openmeteo_get_forecast

Weather forecast for a coordinate pair with hourly and/or daily variable selection.

• Up to 16 forecast days ahead (forecast_days 1–16, default 7)

• past_days (0–92) covers recent history via the forecast model — use instead of openmeteo_get_historical for dates within the last ~5 days to avoid ERA5 lag

• Common hourly variables: temperature_2m, precipitation, wind_speed_10m, relative_humidity_2m, cloud_cover, uv_index, apparent_temperature, precipitation_probability, weather_code, surface_pressure, visibility, wind_direction_10m, wind_gusts_10m, dew_point_2m

• Common daily variables: temperature_2m_max, temperature_2m_min, precipitation_sum, wind_speed_10m_max, sunrise, sunset, uv_index_max, precipitation_hours, weather_code

• At least one of hourly_variables or daily_variables is required

• Hourly and daily are separate variable sets. A variable Open-Meteo documents under the other cadence is rejected before the request, by name, with the field it belongs in and the same-cadence alternatives (cloud_cover in daily_variables → cloud_cover_max/_mean/_min). Names in neither set are passed upstream unchanged

• Configurable temperature unit (Celsius/Fahrenheit), wind speed unit (km/h, mph, m/s, knots), and precipitation unit (mm/inch)

• Reshapes the API's columnar response into per-timestamp records with a parallel hourly_units / daily_units map

• A wide window (a large past_days plus many hourly variables) spills to DataCanvas when CANVAS_PROVIDER_TYPE=duckdb — output includes canvas_id and truncated: true; query with openmeteo_dataframe_query

openmeteo_get_historical

Historical weather from the ERA5 reanalysis archive, covering 1940 to approximately 5 days ago.

• Requires start_date and end_date (YYYY-MM-DD); ERA5 has a variable ~1–5 day lag

• Same variable vocabulary as openmeteo_get_forecast — past and forecast data are directly comparable on one schema

• At least one of hourly_variables or daily_variables is required

• Hourly and daily are separate variable sets; a variable documented under the other cadence is rejected before the request, by name, with the field it belongs in

• Large date ranges (multi-year hourly queries) spill to DataCanvas when CANVAS_PROVIDER_TYPE=duckdb — output includes canvas_id and truncated: true whenever a result is too large to return inline, which a wide multi-variable pull can be at any row count

• Spill → query workflow: call openmeteo_dataframe_describe with the canvas_id to list tables, then openmeteo_dataframe_query to run SQL SELECT against the staged data

openmeteo_get_marine

Marine wave and ocean conditions for coastal and open-ocean coordinates.

• Up to 8 forecast days (forecast_days 1–8, upstream default 7) with optional past_days (0–92)

• Or an archive range via start_date and end_date — real wave values go back to at least 2022

• One window per call: a date range is mutually exclusive with forecast_days/past_days, and needs both ends — a lone start_date or end_date is rejected

• Common hourly variables: wave_height, wave_direction, wave_period, wind_wave_height, wind_wave_direction, wind_wave_period, swell_wave_height, swell_wave_direction, swell_wave_period

• Common daily variables: wave_height_max, wave_direction_dominant, wave_period_max

• At least one of hourly_variables or daily_variables is required

• Hourly and daily are separate variable sets; a variable documented under the other cadence is rejected before the request, by name, with the field it belongs in

• Inland or sheltered-water points return near-zero wave values (physically correct); ocean_current_velocity is null for non-open-ocean coordinates

• Wide windows spill to DataCanvas when CANVAS_PROVIDER_TYPE=duckdb — output includes canvas_id and truncated: true; query with openmeteo_dataframe_query

openmeteo_get_air_quality

Modeled CAMS air quality, forecast and archive.

• Up to 7 forecast days (forecast_days 1–7, upstream default 5) with optional past_days (0–92)

• Or an archive range via start_date and end_date — real CAMS values go back to at least 2022-10-01; earlier dates return rows of nulls

• One window per call: a date range is mutually exclusive with forecast_days/past_days, and needs both ends — a lone start_date or end_date is rejected

• Common variables: pm2_5, pm10, carbon_monoxide, nitrogen_dioxide, sulphur_dioxide, ozone, dust, european_aqi, us_aqi, alder_pollen, birch_pollen, grass_pollen, mugwort_pollen, olive_pollen, ragweed_pollen

• At least one variable from hourly_variables is required

• Grid-modeled data from CAMS — resolution is coarser than ground stations; for measured station readings, cross-reference openaq-mcp-server

• Output includes data_source: "CAMS" to distinguish modeled from measured data

• Wide windows spill to DataCanvas when CANVAS_PROVIDER_TYPE=duckdb — output includes canvas_id and truncated: true; query with openmeteo_dataframe_query

openmeteo_get_elevation

Terrain elevation from the Copernicus Digital Elevation Model (~90m resolution).

• Accepts parallel latitudes[] and longitudes[] arrays; both must have equal length (up to 100 pairs)

• Returns results in input order: { latitude, longitude, elevation_m }

• Useful for geographic context, elevation-adjusted weather interpretation, or route planning

openmeteo_get_ensemble

Probabilistic ensemble weather forecast exposing all individual model member trajectories.

• Up to 16 forecast days (forecast_days 1–16, default 7) with optional past_days (0–92)

• Each requested variable is returned as per-member columns: temperature_2m_member01, temperature_2m_member02, … Use the spread across members to compute exceedance probabilities, interquantile ranges, and decision thresholds

• Available ensemble models (member counts include the control run):

  • Global — ecmwf_ifs025_ensemble (51), ecmwf_aifs025_ensemble (51), google_weathernext2_ensemble (64), ncep_gefs_seamless (31), ncep_gefs025 (31), ncep_gefs05 (31, 35-day horizon), ncep_aigefs025 (31), icon_seamless_eps (20–40, global/Europe blend), icon_global_eps (40), gem_global_ensemble (21), bom_access_global_ensemble (18), ukmo_global_ensemble_20km (18)

  • Regional — ecmwf_ifs_europe_ensemble (51), ecmwf_aifs_europe_ensemble (51), icon_eu_eps (40), icon_d2_eps (20), meteoswiss_icon_ch2_ensemble (21), meteoswiss_icon_ch1_ensemble (11), ukmo_uk_ensemble_2km (3). A regional model returns no data outside the area it covers. Upstream reports that two ways — No data is available for this location from the meteoswiss_* pair, an HTTP 200 carrying nan coordinates from the rest — and both surface as a non-retryable input error naming the coverage gap, so switch to a global model rather than retrying

• Omit models to use the API default blend. The list is not an allowlist — a model name it does not carry is still sent upstream, so a model Open-Meteo adds later keeps working

• Response includes model (system used) and member_count (perturbed members, excluding the control run)

• At least one of hourly_variables or daily_variables is required

• Hourly and daily are separate variable sets; a variable documented under the other cadence is rejected before the request, by name, with the field it belongs in. The ensemble API's own catalog applies — it publishes temperature_2m_max and temperature_2m_min as 3-hourly aggregations as well as daily, so those are accepted in either field

• Large multi-member, multi-day pulls spill to DataCanvas when CANVAS_PROVIDER_TYPE=duckdb — output includes canvas_id and truncated: true; query with openmeteo_dataframe_query

• Configurable temperature, wind speed, and precipitation units

openmeteo_get_flood

GloFAS (Global Flood Awareness System) river discharge forecast and reanalysis via the Open-Meteo Flood API.

• Coordinate-based — no river ID needed; the API snaps to the nearest river grid point automatically

• Forecast horizon up to 210 days; reanalysis history from 1984-01-01 to present

• One mode per call: forecast_days for the future outlook, or start_date and end_date together for historical analysis. The two are mutually exclusive, and a date range needs both ends — a lone start_date or end_date is rejected

• Available daily variables: river_discharge (ensemble mean), river_discharge_mean, river_discharge_min, river_discharge_max, river_discharge_median, river_discharge_p25 (25th percentile), river_discharge_p75 (75th percentile) — all in m³/s

• Returns null for coordinates outside GloFAS coverage (e.g., open ocean or areas without river network data)

• Discharge values reflect the GloFAS ensemble — percentile variables expose the uncertainty spread

• Wide reanalysis ranges spill to DataCanvas when CANVAS_PROVIDER_TYPE=duckdb — output includes canvas_id and truncated: true; query with openmeteo_dataframe_query

openmeteo_get_climate

Long-range climate projections from bias-corrected daily CMIP6 models — the future-projection counterpart to openmeteo_get_historical.

• Coverage: 1950-01-01 to 2050-12-31, daily resolution only

• Available models: CMCC_CM2_VHR4, FGOALS_f3_H, HiRAM_SIT_HR, MRI_AGCM3_2_S, EC_Earth3P_HR, MPI_ESM1_2_XR, NICAM16_8S. Not an allowlist — an unlisted name is still sent upstream; when upstream rejects a multi-model request, the error names only the model outside the documented set, not the whole list

• With 2+ models, each variable appears once per model with the model name as column suffix (e.g. temperature_2m_max_CMCC_CM2_VHR4); a single or omitted model returns plain variable names

• Common daily variables: temperature_2m_max, temperature_2m_min, temperature_2m_mean, precipitation_sum, rain_sum, snowfall_sum, wind_speed_10m_mean, wind_speed_10m_max, shortwave_radiation_sum, cloud_cover_mean, relative_humidity_2m_mean, pressure_msl_mean

• Not all models carry all variables — missing combinations return null (e.g. CMCC_CM2_VHR4 has no shortwave_radiation_sum)

• Multi-decade daily pulls across several models spill to DataCanvas when CANVAS_PROVIDER_TYPE=duckdb — output includes canvas_id and truncated: true; query with openmeteo_dataframe_query

• Configurable temperature, wind speed, and precipitation units

Related MCP server: weather-mcp

Features

Built on @cyanheads/mcp-ts-core:

• Declarative tool definitions — single file per tool, framework handles registration and validation

• Unified error handling — handlers throw, framework catches, classifies, and formats

• Pluggable auth: none, jwt, oauth

• Swappable storage backends: in-memory, filesystem, Supabase, Cloudflare KV/R2/D1

• Structured logging with optional OpenTelemetry tracing

• STDIO and Streamable HTTP transports

Open-Meteo–specific:

• No API key required for non-commercial use — zero-config out of the box

• Self-contained geocoding: openmeteo_search_locations resolves place names so agents don't need a separate geocoder

• ERA5 archive from 1940 to present with same variable schema as the forecast API — direct past/forecast comparisons on one schema

• Automatic columnar-to-record reshape: Open-Meteo returns parallel time/variable arrays; handlers convert to per-timestamp records with a *_units map

• DataCanvas spillover for openmeteo_get_forecast, openmeteo_get_historical, openmeteo_get_marine, openmeteo_get_air_quality, openmeteo_get_ensemble, openmeteo_get_flood, and openmeteo_get_climate: a result too large to return inline registers a DuckDB dataframe for SQL querying, staging every hourly and daily row with its upstream numeric type intact. With CANVAS_PROVIDER_TYPE=none (the default) the same size check still applies — those tools return a bounded preview with truncated: true and no canvas_id, never an unbounded payload claiming to be complete

• Configurable base URLs for all eight API endpoints (forecast, archive, marine, air quality, geocoding, ensemble, flood, climate) — override for testing or self-hosted deployments

• Attribution: Weather data by Open-Meteo.com (CC BY 4.0). Non-commercial use is free and keyless; commercial use requires Open-Meteo's paid API tier (~10,000 req/day, 5,000/hour fair-use ceiling for non-commercial)

Agent-friendly output:

• Location-first workflow: openmeteo_search_locations returns the IANA timezone alongside coordinates — pass it directly as timezone to any weather tool

• Recovery hints on all error contracts — invalid variable names surface correction guidance with common variable examples

• Cadence-aware variable validation on openmeteo_get_forecast, openmeteo_get_historical, openmeteo_get_marine, and openmeteo_get_ensemble: a variable documented under the opposite cadence is rejected before the upstream call, naming the exact value and the field it belongs in, so the next attempt converges instead of re-guessing against an error that echoes the whole requested list. This is not an allowlist — a name in neither documented set goes upstream untouched

• Unserved-variable notice on all seven weather tools: Open-Meteo answers a variable name it parses but does not serve with an all-null column and the unit "undefined" rather than an error, so the result carries a notice naming those columns instead of presenting them as a data gap. openmeteo_get_air_quality, openmeteo_get_flood, and openmeteo_get_climate take a single cadence bucket, so they carry the notice without a cadence guard

• Coordinate snapping transparency — responses echo the snapped latitude/longitude (Open-Meteo quantizes to the nearest model grid point) so agents can reason about grid alignment

• data_source: "CAMS" label on air quality results distinguishes modeled data from measured station readings

Getting started

Public Hosted Instance

A public instance is available at https://open-meteo.caseyjhand.com/mcp — no installation required. Point any MCP client at it via Streamable HTTP:

{
  "mcpServers": {
    "open-meteo-mcp-server": {
      "type": "streamable-http",
      "url": "https://open-meteo.caseyjhand.com/mcp"
    }
  }
}

Self-Hosted / Local

Add the following to your MCP client configuration file.

{
  "mcpServers": {
    "open-meteo-mcp-server": {
      "type": "stdio",
      "command": "bunx",
      "args": ["@cyanheads/open-meteo-mcp-server@latest"],
      "env": {
        "MCP_TRANSPORT_TYPE": "stdio",
        "MCP_LOG_LEVEL": "info"
      }
    }
  }
}

Or with npx (no Bun required):

{
  "mcpServers": {
    "open-meteo-mcp-server": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@cyanheads/open-meteo-mcp-server@latest"],
      "env": {
        "MCP_TRANSPORT_TYPE": "stdio",
        "MCP_LOG_LEVEL": "info"
      }
    }
  }
}

Or with Docker:

{
  "mcpServers": {
    "open-meteo-mcp-server": {
      "type": "stdio",
      "command": "docker",
      "args": ["run", "-i", "--rm", "-e", "MCP_TRANSPORT_TYPE=stdio", "ghcr.io/cyanheads/open-meteo-mcp-server:latest"]
    }
  }
}

For Streamable HTTP, set the transport and start the server:

MCP_TRANSPORT_TYPE=http MCP_HTTP_PORT=3010 bun run start:http
# Server listens at http://localhost:3010/mcp

Prerequisites

• Bun v1.3.0 or higher (or Node.js v24+).

• No API key required. Non-commercial use is free and keyless.

• Commercial use requires Open-Meteo's paid API tier.

Installation

1. Clone the repository:

git clone https://github.com/cyanheads/open-meteo-mcp-server.git

2. Navigate into the directory:

cd open-meteo-mcp-server

3. Install dependencies:

bun install

Configuration

All configuration is validated at startup via Zod schemas. No API key is required for non-commercial use — all variables are optional.

See .env.example for the full list of optional overrides.

Running the server

Local development

• Build and run the production version:

  # One-time build
  bun run rebuild
  
  # Run the built server
  bun run start:http
  # or
  bun run start:stdio

• Run checks and tests:

  bun run devcheck  # Lint, format, typecheck, security
  bun run test      # Vitest test suite

Docker

docker build -t open-meteo-mcp-server .
docker run --rm -p 3010:3010 open-meteo-mcp-server

The Dockerfile defaults to HTTP transport, stateless session mode, and logs to /var/log/open-meteo-mcp-server. OpenTelemetry peer dependencies are installed by default — build with --build-arg OTEL_ENABLED=false to omit them.

Project structure

Development guide

See CLAUDE.md for development guidelines and architectural rules. The short version:

• Handlers throw, framework catches — no try/catch in tool logic

• Use ctx.log for request-scoped logging, ctx.state for tenant-scoped storage

• Register new tools in the tools[] array in src/index.ts

• Wrap external API calls: validate raw → normalize to domain type → return output schema; never fabricate missing fields

Contributing

Issues and pull requests are welcome. Run checks and tests before submitting:

bun run devcheck
bun run test

License

Apache-2.0 — see LICENSE for details.

Weather data by Open-Meteo.com — licensed CC BY 4.0.