bc-observatory 0.4.0 → 0.5.0

package/dist/server.js

@@ -393,9 +393,10 @@ function normOf(r) {
   const symbol = r.kind !== "page";
   const main = /* @__PURE__ */ new Set([...indexTokens(r.name, symbol), ...indexTokens(r.title, symbol), ...r.caption ? indexTokens(r.caption) : []]);
   const own = r.type === "topic" ? new Set(tokenize(r.title)) : null;
-  const tags = (r.tags ?? []).flatMap((t) => indexTokens(t)).filter((t) => !own?.has(t));
+  const learn = r.type === "learn";
+  const tags = (r.tags ?? []).flatMap((t) => indexTokens(t, learn)).filter((t) => !own?.has(t));
   const eqs = r.type === "object" || symbol ? [tokenize(r.name), ...r.caption ? [tokenize(r.caption)] : []] : [tokenize(r.title)];
-  const n = { main: [...main], tags: [...new Set(tags)], text: symbol ? null : indexTokens(r.text ?? ""), eqs };
+  const n = { main: [...main], tags: learn ? tags : [...new Set(tags)], text: symbol ? null : indexTokens(r.text ?? ""), eqs };
   NORM.set(r, n);
   return n;
 }
@@ -783,6 +784,26 @@ function hints(index2, q, hits) {
   return out;
 }
 
+// ../search/src/learn.ts
+function learnRecords(doc) {
+  return doc.rows.map(([path, title, summary, date, h, system, ids]) => {
+    const hub = h >= 0 ? doc.hubs[h] : void 0;
+    const sys = system ?? hub?.[2] ?? null;
+    return {
+      path: /^https?:\/\//.test(path) ? path : `${doc.base}${path}`,
+      type: "learn",
+      title,
+      summary,
+      tier: "official",
+      ...sys ? { system: sys } : {},
+      ...date ? { date } : {},
+      ...ids?.length ? { tags: ids } : {},
+      ...hub?.[1] ? { path_label: hub[1] } : {},
+      ...hub?.[0] ? { hub: hub[0] } : {}
+    };
+  });
+}
+
 // src/embed.ts
 var MAX_TOKENS = 512;
 function parseModel(tokenizerJson, safetensors, config = {}) {
@@ -1243,7 +1264,7 @@ ${pageView(md, clean, {}, budget, site)}`;
 }
 
 // src/server.ts
-var VERSION = "0.4.0";
+var VERSION = "0.5.0";
 var SITE = (process.env.BC_OBSERVATORY_SITE ?? "https://waldo1001.github.io/waldo.BCObservatory/").replace(/\/?$/, "/");
 var LOCAL = process.env.BC_OBSERVATORY_LOCAL ? resolve(process.env.BC_OBSERVATORY_LOCAL) : null;
 var BUDGET = budgetOf(process.env.BC_OBSERVATORY_MAX_CHARS);
@@ -1287,6 +1308,12 @@ async function loadIndex() {
   for (const s of manifest.shards) all.push(...JSON.parse(await shard(s.file, s.sha256)));
   pages = all;
   const idx = prepare(all.map(pageToRecord));
+  if (manifest.learn) {
+    try {
+      prepare(learnRecords(JSON.parse(await shard(manifest.learn.file, manifest.learn.sha256))).map(pageToRecord), idx);
+    } catch {
+    }
+  }
   let sm = null;
   try {
     sm = JSON.parse(await dataFile("index/symbols-manifest.json"));
@@ -1314,7 +1341,10 @@ var line = (r, tag = "") => `- ${r.title}${r.caption ? ` (captioned "${r.caption
 var text = (t) => ({ content: [{ type: "text", text: t }] });
 var symbolHit = (r, tag = "") => `- ${r.name} (${symbolLine(r).replace(/ · /g, ", ")}) path=${r.id}${tag}${r.text ? `
   ${r.text}` : ""}`;
-var hitLine = (r, tag = "") => r.kind === "page" && r.page ? line(r.page, tag) : symbolHit(r, tag);
+var learnLine = (r, tag = "") => `- ${r.title} [learn, official${r.date ? `, ${r.date}` : ""}] url=${r.path}${tag}${r.path_label ? `
+  in: ${r.path_label}${r.hub ? ` (hub: ${r.hub})` : ""}` : ""}
+  ${r.summary}`;
+var hitLine = (r, tag = "") => r.kind === "page" && r.page ? r.type === "learn" ? learnLine(r.page, tag) : line(r.page, tag) : symbolHit(r, tag);
 var MODEL = {
   repo: "minishlab/potion-base-8M",
   revision: "bf8b056651a2c21b8d2565580b8569da283cab23",
@@ -1414,7 +1444,8 @@ async function toolSearch(a) {
 ${hs.map((h) => `Hint: ${h.text}${h.query ? ` (search "${h.query}")` : h.path ? ` (cat("${h.path}"))` : ""}`).join("\n")}` : "";
   if (!shown.length) return `No pages match "${a.query}". Try fewer words, or ls("") to browse sections.${note}${hintText}`;
   const how = semantic.length && keyword.length && mode === "hybrid" ? "keyword + meaning" : mode === "semantic" && semantic.length ? "meaning" : "keyword";
-  return `${shown.length} results, ${how}${note}${how === "keyword + meaning" || mode === "hybrid" ? ` (exact band: ${exactBand})` : ""} (read one with cat(path)):
+  const learnNote = shown.some((r) => r.type === "learn") ? "; a Learn article's url opens on learn.microsoft.com, cat(hub) reads its hub" : "";
+  return `${shown.length} results, ${how}${note}${how === "keyword + meaning" || mode === "hybrid" ? ` (exact band: ${exactBand})` : ""} (read one with cat(path)${learnNote}):
 ${shown.map((r) => hitLine(r, tag(r.id))).join("\n")}${hintText}`;
 }
 async function toolLs(a) {
@@ -1535,13 +1566,13 @@ ${a.message}
 var VIEW = { section: z.string().optional(), offset: z.number().int().min(0).max(1e5).optional(), limit: z.number().int().min(1).max(5e3).optional() };
 function createServer() {
   const server = new McpServer({ name: "bc-observatory", version: VERSION }, {
-    instructions: "BC Observatory: an agent-first knowledge base of Microsoft Dynamics 365 Business Central (Learn hubs, AL objects from the code for BC28-30, localizations, roadmap features, videos, community posts, code changes: merged pull requests of Microsoft's Business Central repositories, BCApps joined to the AL objects they changed, under changes/, and AL Language extension releases (the marketplace changelog, under releases/)). Every page carries a trust tier (official = Microsoft, community = everyone else) and a review state: say which tier a claim comes from. Never invent AL object ids or version numbers: look them up with get_object. Start with search(), read pages with cat(path), browse with ls(path)."
+    instructions: "BC Observatory: an agent-first knowledge base of Microsoft Dynamics 365 Business Central (Learn hubs, AL objects from the code for BC28-30, localizations, roadmap features, videos, community posts, code changes: merged pull requests of Microsoft's Business Central repositories, BCApps joined to the AL objects they changed, under changes/, and AL Language extension releases (the marketplace changelog, under releases/)). Every page carries a trust tier (official = Microsoft, community = everyone else) and a review state: say which tier a claim comes from. Never invent AL object ids or version numbers: look them up with get_object. Start with search(), read pages with cat(path), browse with ls(path). search() also finds Microsoft Learn articles by title, summary and the identifiers they document (type learn): those are not pages here, open their url on learn.microsoft.com."
   });
   server.registerTool(
     "search",
     {
       title: "Search the knowledge base",
-      description: "Search every page, and the fields, published events, global procedures and enum values of AL objects, ranked like the site's search. Takes names ('Sales Header'), references ('t36', 'cu 80', 'table 11300 BE'), bare ids ('36') and symbol prefixes ('event: OnAfterPost', 'field: Posting Date', 'proc: CopyToTempLines'); a symbol hit's path ends in its anchor (#event-<Name>, #field-<id>, #proc-<Name>, #value-<ordinal>); cat(path) reads that member alone. Hybrid by default: exact matches stay on top and the rest is fused with a meaning ranking; each hit says [keyword], [meaning] or [both]; mode 'keyword' or 'semantic' for one list. Filters: type (topic, app, feature, object, localization, video, post, change, source, digest, release), tier (official, community), system (finance, sales, ... or an alias such as 'g/l'), country ('BE'), app ('Subscription Billing'), object_type ('table'), kind (page, field, event, proc, value).",
+      description: "Search every page, and the fields, published events, global procedures and enum values of AL objects, ranked like the site's search. Takes names ('Sales Header'), references ('t36', 'cu 80', 'table 11300 BE'), bare ids ('36') and symbol prefixes ('event: OnAfterPost', 'field: Posting Date', 'proc: CopyToTempLines'); a symbol hit's path ends in its anchor (#event-<Name>, #field-<id>, #proc-<Name>, #value-<ordinal>); cat(path) reads that member alone. Learn articles too: type learn, a url, not a path. Hybrid by default: exact matches stay on top and the rest is fused with a meaning ranking; each hit says [keyword], [meaning] or [both]; mode 'keyword' or 'semantic' for one list. Filters: type (topic, learn, app, feature, object, localization, video, post, change, source, digest, release), tier (official, community), system (finance, sales, ... or an alias such as 'g/l'), country ('BE'), app ('Subscription Billing'), object_type ('table'), kind (page, field, event, proc, value).",
       inputSchema: { query: z.string(), type: z.string().optional(), tier: z.string().optional(), system: z.string().optional(), country: z.string().optional(), app: z.string().optional(), object_type: z.string().optional(), kind: z.enum(["page", "field", "event", "proc", "value"]).optional(), limit: z.number().int().min(1).max(50).optional(), mode: z.enum(["hybrid", "keyword", "semantic"]).optional() }
     },
     async (a) => text(await toolSearch(a))

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bc-observatory",
-  "version": "0.4.0",
+  "version": "0.5.0",
   "description": "MCP server for BC Observatory: an agent-first, cross-referenced knowledge base of Microsoft Dynamics 365 Business Central (Learn, code, roadmap, videos, community blogs).",
   "type": "module",
   "bin": {