Tools
The MCP tools superlore exposes — search, get_page, grep, glob, list, navigate, get_component_data — with their arguments and what each returns.
The superlore MCP server (/api/mcp on this deploy) exposes its corpus two ways: read it (the
page body and the typed knowledge nodes) and find anything in it (grep + glob, like a folder).
Each tool returns JSON — never scraped HTML.
| Tool | Arguments | Returns |
|---|---|---|
| search | query, limit? | Ranked page hits across the corpus |
| get_page | path | Frontmatter + body (MDX) + headings + ordered knowledge nodes |
| grep | pattern, flags?, path?, limit? | Regex matches across page bodies — { path, line, text } |
| glob | pattern | Page paths matching a shell-style glob |
| list | kind?, tag?, entityType? | Filtered knowledge nodes across the corpus |
| navigate | target | Outgoing refs + backlinks for a path / node id / ref |
| get_component_data | id | A single component's knowledge face by id |
Each tool
search(query, limit?)
Full-text search across the documentation, ranked over the page body (not just frontmatter).
Returns page hits; title and summary weigh heaviest.
get_page(path)
Fetch a page's full content by path (e.g. /docs/architecture): its frontmatter, the readable
body (the raw MDX when the deploy wires a content reader, otherwise the extracted prose),
the headings outline, and the ordered knowledge nodes — the serialized faces of every
component on the page. Returns { error: "not_found", path } for an unknown path.
grep(pattern, flags?, path?, limit?)
Regex search across every page body, line by line — ripgrep for the corpus. Returns
{ path, line, text } hits. Pass flags (default i), scope with a path glob, and cap with
limit. This is how an agent finds anything by content, the way you'd grep a folder.
glob(pattern)
List page paths matching a shell-style glob — * within a segment, ** across segments, ?
for one character (e.g. everything under /docs/agents). The corpus's file finder.
list({ kind?, tag?, entityType? })
List knowledge nodes across the whole corpus, filtered by node kind, by frontmatter tag, or
by entityType. Use it to enumerate, say, every timeline or every page tagged roadmap.
navigate(target)
Follow relations from a page path, a node id, or an entity ref. Returns the outgoing refs
(depends-on, part-of, related, …) and the backlinks — the corpus as a graph.
get_component_data(id)
Get the typed data behind a single rendered component (its knowledge
face) by id — e.g. the node/edge graph behind a Canvas,
or the items behind a Timeline. Returns { error: "not_found", id } for an unknown id.
The same index the site renders
Every tool reads the structured index superlore builds from your MDX. There is no separate agent content to maintain — the tools and the pages are two projections of one source.
To wire a client up to these tools, see Connect a client.