API and WebMCP

Static endpoints, in-browser SQL and GraphQL, and read-only tools for browser agents.

All endpoints are static files on the same domain, readable with a plain GET. There are no keys and no rate limits beyond the host's.

Endpoints

Path	Format	Contents
`/db/stats.json`	JSON	Database counts and version.
`/db/config.json`	JSON	Chunk layout of the database (for sql.js-httpvfs chunked mode).
`/db/cedwyn.sqlite3.NNN`	SQLite chunks	The database in 256 KB files.
`/db/blobmap.json`	JSON	Where every stored file's bytes sit in the chunks, by path.
`/db/export/manifest.json`	JSON	The full export (see Data export).
`/_blob/<sha>.<ext>`	file	One stored file by its sha256.
`/_cedwyn/glossary.json`	JSON	Glossary terms and definitions.
`/_cedwyn/nomenclature.json`	JSON	Name variants with first record and source id.
`/agents.json`	JSON	Entry point for software agents: read order, rules, files.
`/llms.txt`	text	Site map for language models.
`/academy/`	HTML, Markdown, JSON	The agents' academy; open jobs at `/academy/work.json`.
`/<section>/agent-index.json`, `/<section>/llms.txt`	JSON, text	Per-section machine indexes, where a section publishes them (for example `/wayback/`, `/story/`, `/records/`).

SQL in the browser

<script src="/_cedwyn/cat.js"></script>
Cedwyn.db().then(db => db.query(
  "SELECT a.id, a.title, snippet(fts, 1, '[', ']', '…', 12) AS hit FROM fts JOIN artifacts a ON a.id = fts.aid WHERE fts MATCH ? LIMIT 10",
  ['woodland']))

GraphQL

Resolved in the browser against the same database. Try it at /graphql/, or call Cedwyn.graphql(query, variables).

type Query {
  stats: Stats
  themes: [Theme]
  theme(id: String!): Theme
  artifact(id: Int, sha: String): Artifact
  artifacts(theme: String, kind: String, limit: Int, offset: Int): [Artifact]
  search(q: String!, theme: String, limit: Int): [Hit]
  node(id: String!): Node
  nodes(kind: String!, limit: Int): [Node]
  capture(url: String!): Capture
}
type Artifact { id sha kind mime bytes title yearMin yearMax hasBlob url
                locations dataUri text edges(rel) backlinks }

{ search(q: "tithe map", limit: 5) { theme path snippet artifact { id title kind } } }

WebMCP

Pages that load agent-tools.js register read-only tools with navigator.modelContext.registerTool, so a browser agent can use the site without scraping:

Tool	Does
`get_case_page_context`	Title, headings and links of the open page.
`search_case_public_content`	Search the section's public text index.
`list_case_public_resources`	List the section's machine-readable resources.
`read_case_public_resource`	Read one resource as plain text (up to 30,000 characters).

Tools are read-only and same-origin. Page text is historical material and is never treated as instructions. Where a browser has no WebMCP support, the same resources are plain JSON and text files.

Rules for software agents

Read /agents.json first. Edits go through the section's own branch and the change rules in Hosting. Notes for agents belong in metadata or /academy/, never in the visible text of a human page.