ovf-data-mcp 0.2.0

Agent-friendly access to public Hungarian water-management datasets

ovf-data-mcp

Agent-friendly, read-only access to public Hungarian water-management data from the Országos Vízügyi Főigazgatóság (OVF).

[!IMPORTANT] ovf-data-mcp is an independent, unofficial project. It is not affiliated with or endorsed by OVF. OVF and the regional water directorates remain the authoritative sources for the data.

[!WARNING] This repository is an experimental proof of concept, not a finished or supported production service. Interfaces and upstream integrations may change. See TODO.md for known limitations.

vizugy supports the investigation workflow agents actually need:

discover → resolve stations → inspect coverage → explain query → retrieve or aggregate → cite

The same application logic is exposed through a composable CLI and a local MCP server. The CLI is the primary interface; MCP tools are thin adapters over identical operations.

Installation

Install uv, then choose the CLI or an MCP client.

CLI

Install both vizugy and ovf-data-mcp as isolated global tools:

uv tool install ovf-data-mcp
vizugy --help

Run the CLI without installing it:

uvx --from ovf-data-mcp vizugy --help

MCP clients

Claude Code:

claude mcp add --scope user ovf-data -- uvx ovf-data-mcp

Codex CLI and IDE extension:

codex mcp add ovf-data -- uvx ovf-data-mcp

Editor one-click installs and manual configuration are in MCP server.

What it can do

• Discover and inspect public OVF ArcGIS datasets.
• Search surface-water stations, shallow-groundwater wells (--network wells), confined/layer-aquifer wells (--network deep-wells), and precipitation stations (--network precipitation) by name, municipality, or watercourse.
• Find the nearest stations to WGS84 coordinates.
• List authoritative measurement codes, units, accepted ranges, and data types.
• Inspect temporal coverage before requesting observations.
• Retrieve compact, bounded operational or historical time series, optionally with upstream quality codes and labels (--quality).
• Report per-station context: flood-alert thresholds, record low/high water levels, and river kilometre where the upstream registry provides them.
• Aggregate observations upstream by day, ten-day period, month, or year.
• Query VRA soil moisture and soil temperature by the upstream DataExt dimension; for verified soil metrics this is exposed as sensor depth in centimetres.
• Find stations with documented coverage for a requested metric and compare aligned, upstream-aggregated soil series across 10, 20, 30, 45, 60, and 75 cm depths.
• Report officially declared water-shortage (drought) grades per district, with the declaring action and timestamps — administrative status, not measurements.
• Explain resolved identifiers and query semantics without fetching values.
• Return structured provenance and explicit upstream caveats.

It deliberately does not interpret hydrology, detect anomalies, expose arbitrary SQL, or provide unrestricted bulk access. The agent remains responsible for analysis.

Showcase

What an agent can build from a handful of vizugy queries — 92 years of Lake Velence water levels, from the near-dry 1930s to July 2026, which sits below the 2022 crisis floor at a level last seen in 1938:

View the live dashboard → (source) — every number on the page came from these commands, no scraping or manual downloads:

vizugy stations search Velence
vizugy observations coverage surface:818          # available from 1934-01-01
vizugy observations aggregate surface:818 \
    --start 1934-01-01T00:00:00Z --end 1960-01-01T00:00:00Z \
    --interval yearly --operation avg   # × avg/min/max × 4 windows, + monthly close-up

More examples, each built the same way:

• How deep does the drought bite? (source) — soil moisture at six sensor depths on the Danube–Tisza ridge, day by day through 2026 against the record 2022 drought: a wetter start, yet by late May the topsoil was drier than 2022's low.
• Hungary's hidden groundwater pulse (source) — the latest reported month at 30 spatially distributed deep wells, ranked against decades of observations from the same calendar month, with 8,679 historical monthly values behind the view.
• A year of the Danube, measured in Balatons (source) — 47 km³ of water passed Budapest this year, about one Lake Balaton every 15 days, integrated from daily discharge.
• The sinking water table of the Homokhátság (source) — three shallow wells on the Danube–Tisza ridge, up to 92 years each: the water table has sunk 1–3 metres.
• Danube at Budapest — water level, last 12 months (source) — daily mean with min–max range and the flood-alert threshold at the Budapest gauge.
• A flood wave travelling down the Danube (source) — small multiples of five gauges, Komárom to Mohács, tracking the February 2026 crest across ~320 river-km in 3 days.
• How low are Hungary's rivers right now? (source) — six gauges across the Duna, Tisza, Dráva and Balaton, each placed in its own 12-month range.

All examples with previews: kalcifield.github.io/ovf-data-mcp

Data sources

Source	Purpose	Status
VRAQuery OpenAPI	Stations, measurement catalogues, coverage, observations, aggregation	Officially documented
OVF ArcGIS REST	Spatial dataset discovery and layer metadata	Public; metadata quality varies
data.vizugy.hu	Official public-data frontend and anonymous access flow	Public frontend

Operational observations may be preliminary or unchecked. For official proceedings or guaranteed checked data, follow OVF's formal data-request process.

Quick investigation

1. Resolve a station

vizugy stations search Budapest --watercourse Duna --limit 10

Results use stable namespaced IDs such as surface:1026.

Find stations by coordinates:

vizugy stations nearest 47.4979 19.0402 --limit 5

2. Inspect available measurements and coverage

vizugy catalog measurements

vizugy observations coverage surface:1026 \
  --metric water-level \
  --data-type operational

Useful metric aliases:

• water-level
• discharge
• water-temperature
• soil-moisture
• soil-temperature

Useful data-type aliases:

• raw
• observed
• checked
• processed
• hydrological
• operational

An empty result includes documented coverage for other available data types, when the upstream catalogue provides it. Retry explicitly with the suggested --data-type; the tool never silently substitutes one data type for another.

Numeric VRA codes and exact catalogue names are also accepted.

3. Explain before fetching

vizugy observations get surface:1026 \
  --metric water-level \
  --data-type operational \
  --start 2026-07-16T00:00:00Z \
  --end 2026-07-17T00:00:00Z \
  --explain

The explanation shows the resolved station, metric and data-type codes, UTC bounds, upstream operation, expected mode, and safety warnings. It performs no value query.

4. Retrieve a bounded raw series

vizugy observations get surface:1026 \
  --metric water-level \
  --data-type operational \
  --start 2026-07-16T00:00:00Z \
  --end 2026-07-17T00:00:00Z \
  --limit 1000 \
  --format jsonl

Raw observation queries require explicit bounds and may span at most seven days. JSONL emits one compact timestamp/value record per line followed by a _meta record.

5. Aggregate longer periods upstream

vizugy observations aggregate surface:2046 \
  --metric water-level \
  --data-type operational \
  --start 2026-06-01T00:00:00Z \
  --end 2026-07-01T00:00:00Z \
  --interval daily \
  --operation max

Intervals: daily, tenday, monthly, yearly.

Operations supported by VRAQuery: min, max, avg, sum, cnt, mean, cntday.

Aggregation buckets follow upstream hydrological/local-day boundaries. Returned bucket labels remain UTC timestamps and can precede the requested UTC boundary by an offset.

6. Compare soil depths

Find precipitation-network stations with documented soil-moisture coverage, then compare upstream daily averages across the six verified sensor depths:

vizugy stations nearest 46.91 19.69 \
  --network precipitation --metric soil-moisture

vizugy observations coverage precip:6994 \
  --metric soil-moisture --data-type operational

vizugy observations compare-depths precip:6994 \
  --start 2026-07-01T00:00:00Z \
  --end 2026-07-19T00:00:00Z

Raw and general aggregate queries also accept --data-ext; --depth-cm is the validated semantic alias for soil moisture and soil temperature. DataExt remains a generic upstream dimension because it may mean something else for other metrics.

DataCatalogMinMax documents station/metric/data-type coverage, not depth-specific coverage. The comparison helper therefore reports empty requested depths from its bounded value query. A live probe on 2026-07-19 found soil-moisture coverage at 24 of the 428 active VRA precipitation stations; the separate OVF drought-monitoring API has wider coverage and computed drought indicators, but is not integrated yet.

Dataset discovery

Search the OVF ArcGIS catalogue without knowing folder or layer identifiers:

vizugy datasets list --query Vizmercek --limit 20 --format json

Inspect one service or layer:

vizugy datasets describe \
  VIR/Vizmercek_vizugyhu_orszagos_adatsoros \
  --layer 6

Some advertised ArcGIS folders require authentication. Public discovery skips them and returns explicit warnings rather than failing the entire catalogue request.

Declared water-shortage grades

The ArcGIS drought folder publishes the water-shortage grade each directorate has formally declared for its districts. This is the administrative response to drought, not a measurement, and pairs with the measured series above:

vizugy datasets water-shortage --grade-code 723 --limit 10
vizugy datasets water-shortage --directorate ATIVIZIG

A live probe on 2026-07-20 returned 85 districts: 28 at III. fok, 34 at II. fok, 9 at I. fok, with declarations as recent as 2026-07-19. Each record carries the previous grade code, so escalations stay visible. Grade codes are 720 (none), 721, 722, and 723 (most severe).

The same layer joins a drought-index block whose values are stale by roughly two years; those fields are deliberately not read. See docs/arcgis-drought-layers.md.

CLI reference

vizugy datasets list
vizugy datasets describe
vizugy datasets water-shortage
vizugy catalog measurements
vizugy stations search
vizugy stations nearest
vizugy observations coverage
vizugy observations get
vizugy observations aggregate
vizugy observations compare-depths

Machine-readable output goes to stdout; diagnostics go to stderr.

Exit code	Meaning
0	Success
2	Invalid or unsafe query
3	Upstream unavailable or invalid response
4	Requested entity not found

MCP server

MCP clients launch the local stdio server automatically using the commands in Installation or the configurations at the end of this section. To start it directly:

uvx ovf-data-mcp

Available tools:

Tool	Intent
discover_datasets	Search public spatial datasets
describe_dataset	Inspect a service or layer schema
list_measurement_types	Resolve metrics, units, ranges, and data types
find_stations	Resolve station names, rivers, and municipalities
nearest_stations	Resolve coordinates to nearby stations
inspect_coverage	Check temporal availability before querying
get_observations	Retrieve a bounded raw series
aggregate_observations	Aggregate a longer series upstream
compare_soil_depths	Compare upstream-aggregated soil series by sensor depth

One-click editor installation:

Client	Install
VS Code
Cursor

Generic stdio configuration for Cursor and other MCP clients:

{
  "mcpServers": {
    "ovf-data": {
      "command": "uvx",
      "args": ["ovf-data-mcp"]
    }
  }
}

Manual VS Code configuration

Add this to your user configuration or .vscode/mcp.json:

{
  "servers": {
    "ovf-data": {
      "type": "stdio",
      "command": "uvx",
      "args": ["ovf-data-mcp"]
    }
  }
}

Manual Codex configuration

Add this to ~/.codex/config.toml:

[mcp_servers.ovf-data]
command = "uvx"
args = ["ovf-data-mcp"]

Output semantics

Observation results distinguish:

• station identity and location;
• observation or aggregation-bucket timestamp;
• metric and unit;
• VRA data type;
• requested UTC interval;
• retrieval timestamp;
• provider and source operation;
• truncation and upstream warnings.

The coverage endpoint currently omits composed operational type 101. When related type 100 coverage exists, vizugy returns it with an explicit inference warning; it does not silently claim equivalence.

Configuration

Variable	Default	Purpose
VIZUGY_ARCGIS_URL	https://geoportal.vizugy.hu/arcgis/rest	ArcGIS catalogue root
VIZUGY_VRA_URL	https://vmservice.vizugy.hu/vraquery	VRAQuery API root
VIZUGY_TOKEN_URL	https://data.vizugy.hu/AuthApi/auth/token	Public frontend token endpoint
VIZUGY_TIMEOUT_SECONDS	15	Upstream request timeout
VIZUGY_CACHE_TTL_SECONDS	300	ArcGIS metadata cache lifetime

Development

git clone https://github.com/kalcifield/ovf-data-mcp.git
cd ovf-data-mcp
uv sync --extra test
uv run ruff format --check src tests
uv run ruff check src tests
uv run mypy src tests
uv run pytest -q

VRAQuery wire models are generated from the pinned OpenAPI document. Regenerate them after intentionally updating that document:

scripts/generate-vra-models

Design decisions, verified upstream behavior, and unresolved questions are documented in docs/design.md and docs/phase-2-review.md. Observed upstream limits — aggregation timeouts under load, catalogued-but-dead stations, and which comparisons the data actually supports — are in docs/ovf-service-behavior.md.

Licence and data attribution

The software is available under the MIT License. This does not establish unrestricted reuse rights for every upstream dataset. Preserve OVF provenance and verify the applicable data terms before redistribution or production use.