For machines
Integrations & endpoints
A standard whose subject is machine-readable honesty should be trivial for a machine to read. Everything here is static, CC0, and served without a request back to us.
Endpoints
/levels.json- The scale itself — levels, definitions, decision tree, mappings. The single source of truth, CC0. Also per-language at /<lang>/levels.json.
/levels.schema.json- JSON Schema for a declaration object, so a tool can validate one without hard-coding the rules.
/ecosystem.json- The AI-provenance ecosystem, mapped: what the scale interoperates with, complements, and does not rely on.
/llms.txt- The map for language models (llmstxt.org) — levels, decision procedure, docs.
/llms-full.txt- The whole standard in one file: manifesto, levels, edge cases, mappings. One fetch, no crawl.
/feed.xml- RSS of releases. A change to what a level means invalidates declarations made against it — subscribe to know when the standard moves.
/.well-known/ai-disclosure.json- This site's own declaration, at a well-known path.
/sitemap-index.xml- Every page and its translations, with reciprocal hreflang.
MCP server
An MCP server so an AI agent can query the scale directly — classify a work through the five questions, look up a level, or read the spec — instead of scraping the site. It runs locally over stdio and needs no network: the whole standard travels inside the package.
Add it to an MCP client (e.g. an mcp.json config)
Tools: classify, get_level, list_levels, get_spec. Resources: levels.json, llms-full.txt.
{
"mcpServers": {
"usagescale": {
"command": "npx",
"args": ["-y", "usagescale-mcp"]
}
}
}Put the declaration in your <head>
Each level page carries a copy-paste block. In short: <meta name="ai-usage"
content="3"> plus the experimental ai-disclosure value and a
schema.org term. See any level, e.g. /3.