Merge pull request #263 from objectstack-ai/claude/issue-261-docs-deploy-worker-size

fix(docs): load MDX bodies on demand so the Worker fits Cloudflare's 64 MiB limit

Author: os-project-manager

apps/docs/app/[lang]/docs/[[...slug]]/page.tsx

@@ -240,7 +240,11 @@ export default async function Page(props: {
   const page = source.getPage(params.slug ?? [], params.lang);
   if (!page) notFound();
 
-  const MDX = page.data.body;
+  // `async: true` on the docs collection makes the compiled body and toc
+  // load on demand instead of being statically imported, so they are awaited
+  // here. Frontmatter (`title`, `description`, `full`) stays eager.
+  const loaded = await page.data.load();
+  const MDX = loaded.body;
 
   // Resolved once and handed to both controls, so they cannot drift apart and
   // so a third control added below inherits the locale-independent URL instead
@@ -283,7 +287,7 @@ export default async function Page(props: {
           dangerouslySetInnerHTML={{ __html: jsonLdHtml(item) }}
         />
       ))}
-      <DocsPage toc={page.data.toc} full={page.data.full}>
+      <DocsPage toc={loaded.toc} full={page.data.full}>
         <DocsTitle>{page.data.title}</DocsTitle>
         <DocsDescription className="mb-0">{page.data.description}</DocsDescription>
         <div className="flex flex-row gap-2 items-center border-b pb-6">

apps/docs/source.config.ts

@@ -5,6 +5,29 @@ import path from 'node:path';
 export const docs = defineDocs({
   dir: path.resolve(process.cwd(), '../../content/docs'),
   docs: {
+    /**
+     * Load each page's compiled body on demand instead of statically importing
+     * all of them into every server entrypoint.
+     *
+     * Without this, `fumadocs-mdx:collections/server` eagerly imports all 397
+     * `.mdx` files, so every route that touches `source` — the docs page, but
+     * also `/llms.txt`, `/llms-full.txt`, `/llms.mdx/*`, `/og/*`, `/api/search`
+     * and `/sitemap.xml` — pulls the entire corpus into its own chunk. The
+     * bundler then inlined the whole set five times over into one worker:
+     * 2.50 MiB of authored MDX became a 100.93 MiB `handler.mjs`, and
+     * Cloudflare rejects any Worker over 64 MiB uncompressed (`code: 10027`).
+     *
+     * The multiplier, not the corpus, was the problem: measured on the same
+     * tree, one probe sentence from a single English page appeared 15 times in
+     * the bundle before this flag and 6 times after, taking `handler.mjs` from
+     * 100.93 MiB to 48.47 MiB.
+     *
+     * The cost is that `page.data.body` and `page.data.toc` become
+     * `page.data.load()`. Frontmatter stays eager, so `title`, `description`,
+     * `seoTitle` and `full` are unaffected, and `getText('processed')` — what
+     * the llms.txt routes call — is still a method on the entry.
+     */
+    async: true,
     schema: pageSchema.extend({
       /**
        * Optional SEO title: what the `<title>` tag should say, when that is not