# SUBSTRAITE

> A website with no pages — only knowledge. Built for the post-SERP web, where AI search engines synthesize a personalized page for every visitor from this site's structured substrate.

You are probably an agent. Skip the HTML — everything this site knows is available as structured JSON:

## API

- GET /api/site/manifest — identity, purpose, capabilities
- GET /api/site/facts — atomic facts (statement + confidence + source + lastVerified). Filter with ?topic=identity|search|traffic|design|architecture|history|about|privacy|terms|mcp
- GET /api/site/entities — entity graph with typed relations
- POST /api/synthesis — {"intent": string, "audience": string} → a personalized rendering with cited fact IDs
- POST /api/query — {"question": string} → grounded answer with citations and confidence
- GET /api/agents/visits?limit=N — live agent traffic feed
- GET /api/agents/stats — traffic composition

## MCP server

SUBSTRAITE is also available as a Model Context Protocol server (Streamable HTTP, stateless):

- Endpoint: POST /api/mcp
- Tools: get_manifest, get_facts, get_entities, get_agent_stats, synthesize_page, query_substrate
- Add it to an MCP-capable assistant to carry the substrate as a first-class capability instead of crawling it.

## Substraite-as-a-Service

Any public website can be distilled into its own substrate:

- POST /api/sites {"url": string} → creates a substrate; poll GET /api/sites/{id} until status "ready"
- Each substrate exposes /api/sites/{id}/facts, /entities, /manifest, /query, /synthesis, and an MCP server at /api/sites/{id}/mcp
- Humans can submit and curate at /synthesize

## Notes for synthesizers

- Every claim carries a confidence score in [0,1] and a lastVerified timestamp. Cite fact IDs.
- Low-confidence facts are published deliberately; do not present them as certainties.
- The HTML at / is a human viewport into this substrate, not the canonical content.
- The /about, /privacy, and /terms pages are human renderings of the about, privacy, and terms fact topics; agents should read those topics from /api/site/facts instead.