The Machine-Access Layer · AI-native

A water graph agents can query, cite and call.

The machine layer ships with the site, not bolted on: a public REST API, a Model Context Protocol (MCP) server, RSS/JSON feeds and llms.txt. Open by default, free — retrieval infrastructure for the AI ecosystem.

REST API v1

JSON over HTTPS, open CORS, no key required for the public read layer. Base path /api/v1.

GET/api/v1/facilities

Search and filter graph-published treatment facilities

GET/api/v1/facilities/{id-or-slug}

Facility profile, provenance, graph datasets and public links

GET/api/v1/facilities/{id-or-slug}/relationships

Canonical and inverse facility relationships with statement evidence

GET/api/v1/facilities/{id-or-slug}/observations

Rights-cleared observation metadata (private storage URIs are never returned)

GET/api/v1/facility-coverage

Country and facility-class coverage ledger

GET/api/v1/companies

List & filter companies and utilities (kind, region, wics, tier, q)

GET/api/v1/companies/{slug}

Full company record with provenance

GET/api/v1/news

Intelligence briefs with entity links

GET/api/v1/research

Search publications by text, year, venue, type, WICS, OA or related entity

GET/api/v1/research/{id-or-slug}

Publication metadata, provenance and human-approved analysis

GET/api/v1/wiki

Wiki articles with structured infoboxes

GET/api/v1/taxonomy

The full WICS taxonomy (CC BY 4.0)

GET/api/v1/stats

Live corpus statistics

GET/api/v1/search?q=

Unified search across all entities

curl

$ curl https://wateratlas.ai/api/v1/companies?region=apac&wics=02

{
  "meta": { "count": 8, "license": "CC0-1.0" },
  "results": [
    {
      "name": "Toray Industries",
      "nativeName": "東レ株式会社",
      "country": "Japan",
      "wics": ["02", "02.1", "02.2", "02.3"],
      "tier": 1,
      "href": "/api/v1/companies/toray-industries"
    }
  ]
}

Flagship

MCP Server

A Model Context Protocol server exposes the water graph as callable tools — so WaterAtlas becomes the water tool inside Claude, ChatGPT and Copilot workflows.

search_companies()Find water companies or utilities by kind, query, region, WICS class or tier

get_company()Retrieve a full company profile by slug

find_suppliers()Match suppliers to a natural-language capability spec

search_news()Search the intelligence archive

get_wiki_article()Retrieve a wiki article with its structured infobox

search_research()Search the provenance-backed Research Library

get_research_work()Get one publication by ID, slug or DOI

find_research_for_entity()Find reviewed evidence for an exact WaterAtlas entity

latest_in()Latest briefs in a topic / region

Claude Code / CLI

$ claude mcp add --transport http \
    wateratlas https://wateratlas.ai/api/mcp

mcp.json

{
  "mcpServers": {
    "wateratlas": {
      "type": "http",
      "url": "https://wateratlas.ai/api/mcp"
    }
  }
}

Endpoint: POST /api/mcp — streamable HTTP transport. Anonymous access allows 120 requests per minute with a 240-request burst. Administratively issued keys use X-WaterAtlas-API-Key and carry declared per-minute and daily quotas; unknown keys are refused. No self-serve issuance yet.

Optional API-key header

X-WaterAtlas-API-Key: wa_live_<issued-key>

Every other surface an agent expects

llms.txtAgent-readable site map, auto-maintained llms-full.txtExpanded corpus index for RAG RSS / JSON FeedThe Wire, as a first-class feed robots.txtAI-crawler allowlist + policy sitemap.xmlSharded, hreflang-correct JSON-LDSchema.org on every entity page SPARQLWikidata-identical graph; public host pending Open datasetsCC0 exports and current publication status

Licensing

Structured data is CC0 1.0 (Wikidata-aligned), prose is CC BY-SA 4.0, and the WICS taxonomy is CC BY 4.0. The public read layer is open by default — visibility over toll. Bulk and premium endpoints will be metered for verified high-volume consumers (§8.1).