Open-source manufacturing ERP, MES and QMS. Quoting, MRP, inventory, shop floor, quality and lot/serial traceability on one Postgres schema, with a REST API and MCP server. Self-host or use Carbon Cloud.
TypeScript ai-agentsbill-of-materialserpinventory-managementleanmanufacturingmanufacturing-execution-systemmcp-servermesmrpopen-source-erppostgresqlproduction-planningqmsquality-management-systemreact-routerself-hostedsupabasetraceabilitytypescript
// SPDX-License-Identifier: AGPL-3.0-only
// Carbon (github.com/crbnos/carbon). Modified or adapted versions of this file,
// including ports, remain AGPLv3; serving them over a network requires releasing their source.
import { stemInflection } from "@carbon/content/search";
import { createSearchAPI } from "fumadocs-core/search/server";
import { buildSearchIndexes } from "@/lib/search-index";
/* Single search endpoint across all four surfaces (Reference docs, the Guide, API
* resources, MCP tools). Canonical fumadocs pattern: one combined `indexes` array, each
* entry `tag`ged (docs | guide | resources | tools) so the header's surface pills can
* filter via `?tag=`. */
// zbsearch ignores term frequency, so a one-word fragment ("Shipment" in the audit
// log's entity table) scores the same as a page titled "Shipments", and the tie falls
// to insertion order. Weighting each page's own title row puts the page about the
// term ahead of pages that only mention it; groups follow this order.
const TITLE_WEIGHT = 2;
const weighted = ([, score, doc]: [unknown, number, unknown]) =>
(doc as { type?: string }).type === "page" ? score * TITLE_WEIGHT : score;
const { GET: search } = createSearchAPI("advanced", {
language: "english",
// Inflection-only stemming ("purchase order" matches "Purchase orders"), the same
// stemInflection MCP search_tools and the agent's search_docs add to their queries.
tokenizer: { language: "english", stemming: true, stemmer: stemInflection },
search: { sortBy: (a, b) => weighted(b) - weighted(a) },
indexes: buildSearchIndexes()
});
/**
* Fumadocs has no server-side result limit, and with 1,495 operations indexed a
* common prefix is pathological: "get" matches every READ operation and returned a
* 544 KB body — per keystroke — of which the palette displays eight pages.
*
* Results arrive relevance-ordered, so truncating keeps the best matches. The cap is
* well above what the client can show (MAX_PAGES x MAX_ROWS_PER_PAGE in
* `search-command.tsx`) so grouping still has spare rows to work with; it exists to
* stop the tail, not to do the client's job.
*/
const MAX_RESULTS = 80;
export async function GET(request: Request): Promise<Response> {
const response = await search(request as Parameters<typeof search>[0]);
if (!response.ok) return response;
const results = await response.json();
if (!Array.isArray(results) || results.length <= MAX_RESULTS) {
return Response.json(results);
}
return Response.json(results.slice(0, MAX_RESULTS));
}