FAQ¶
What is AQUAVIEW?¶
AQUAVIEW is a unified discovery layer for ocean, atmospheric, and marine data. It indexes 700,000+ datasets across 68 authoritative sources (NOAA, IOOS, World Ocean Database, Argo, ESA, IFREMER, etc.) into a single STAC catalog and exposes that catalog as a Model Context Protocol (MCP) server so any LLM agent can query it directly.
Platform: https://aquaview.org MCP overview: https://aquaview.org/mcp-overview
Is the MCP server free?¶
The hosted endpoint at https://mcp.aquaview.org/mcp is currently free to use, no key required. Heavy production workloads should reach out via the homepage.
Does AQUAVIEW host the actual data files?¶
No. AQUAVIEW indexes metadata and exposes asset links (hrefs) that point at the authoritative providers — NCEI, AWS Open Data, Google Cloud Storage, ERDDAP, and so on. Data ownership stays with the source institution. AQUAVIEW is the discovery and routing layer.
What's the difference between AQUAVIEW MCP and ERDDAP / STAC API directly?¶
| Aspect | AQUAVIEW MCP | Direct ERDDAP / STAC API |
|---|---|---|
| Sources unified | 68 in one query | Per-server |
| Query language | Natural language → CQL2 | Custom per server |
| LLM-native | Yes (MCP) | No |
| Per-variable filtering | Yes (column_stats_summary) |
Source-dependent |
| Aggregations | Native | Source-dependent |
AQUAVIEW ingests, normalizes, and re-publishes those upstream catalogs into a single schema. You give up some source-specific richness in exchange for unified search across decades of data and dozens of providers.
Which LLMs are supported?¶
Any LLM with MCP support, including:
- Claude (Desktop, Code, Anthropic API direct)
- ChatGPT (Custom Connectors, OpenAI Agents SDK)
- Gemini (Google Gen AI SDK)
- Grok (xAI clients with MCP support)
- Cursor, Cline, Continue, Zed, and other MCP-enabled IDEs
See INSTALL.md for setup per client.
Will my queries be logged?¶
The hosted endpoint records request metadata (timestamps, tool names, parameters) for operational monitoring and abuse prevention. It does not record user-attributable identifiers unless you authenticate. See the privacy notice on aquaview.org for the canonical policy.
What's the rate limit?¶
Currently soft-capped at a generous default per IP. Hard limits aren't published; if you hit one you'll get a 429 with a Retry-After. Open an issue if your legitimate use is being rate-limited.
How fresh is the data?¶
| Source family | Update cadence |
|---|---|
| Realtime buoys (NDBC, NOS_COOPS) | Minutes-to-hourly |
| Satellite (GOES-R, Sentinel) | Hours after acquisition |
| Argo (GADR) | Daily near-real-time + delayed-mode reprocessing |
| Weather model (HRRR) | Hourly |
| Archive (WOD, NEXRAD historical) | Periodic batches |
AQUAVIEW re-indexes on a rolling schedule. Item-level "last modified" is preserved from the source.
Can I add a source?¶
Yes — open an issue with a link to the upstream catalog and a short description. AQUAVIEW prioritizes sources with stable APIs and authoritative provenance.
What does the aquaview: property prefix mean?¶
These are normalized properties AQUAVIEW adds on top of source-specific ones (which retain their original names). The normalization makes cross-source filtering possible. See data-model.md.
Does AQUAVIEW support GraphQL / REST in addition to MCP?¶
This repo focuses on MCP. The underlying STAC API is also exposed; ask on the homepage for access.
My agent is calling search_datasets with q="Gulf of Mexico" and getting weird results¶
That's a known prompting pitfall — region names in free text are unreliable. Use bbox. See prompting-guide.md.
My agent's response includes oil-spill incident reports when I asked about ocean temperatures¶
INCIDENT_NEWS is text-heavy and over-matches free-text searches. Add exclude_collections="INCIDENT_NEWS". See prompting-guide.md.
How do I download the actual NetCDF / GRIB2 / CSV file?¶
Either ask the LLM directly ("give me the download URL for that item") or call get_item(collection, item_id) to receive the full asset list. The href on each asset is a direct download URL at the source provider.
I'm seeing 68 collections but the docs mention "15 sources" elsewhere¶
"Sources" sometimes refers to source organizations (e.g., NOAA), and "collections" refers to the per-program subdivisions (NDBC, COOPS, GOES-R, etc., are all NOAA but are separate collections). The number of collections is the more useful figure for filtering: 68.
How do I cite AQUAVIEW in academic work?¶
Cite the underlying data source — AQUAVIEW is a discovery layer, not a primary publisher. AQUAVIEW's catalog records preserve aquaview:institution, original DOIs where available, and asset URLs back to the authoritative provider. We recommend citing both: the original dataset, and AQUAVIEW as the discovery tool ("dataset discovered via aquaview.org").