@strand-cms/core

The core of Strand, an agent-first publishing system: the validated content schema, the MDX-in-Git loader, and the SEO/GEO generators every Strand site emits. No database, no CMS UI — posts are MDX files with strict frontmatter, and everything here is a pure function over them.

npm install @strand-cms/core

Schema (@strand-cms/core/schema)

Zod schemas that define the content contract agents write against:

• PostFrontmatter — the post contract: slug, title, description (50–160 chars), status; SiteConfig.description is 25–160 (draft/scheduled/published), tags, authors, SEO fields, cited sources, FAQ, and AI-search targeting: contentType (guide/comparison/roundup/news/ explainer), primaryKeyword, and per-article keywords (5–8).
• AuthorFrontmatter — author entities for E-E-A-T markup.
• SiteConfig / RoutesConfig — site identity and URL layout, with postPath, postUrl, tagPath, authorPath helpers, and the opt-in generateOgImages flag for theme-generated per-post social cards (off by default — a post with no image source still publishes, just without og:image).
• SourcePolicy + checkSourcePolicy(sources, policy) — mechanical enforcement of editorial sourcing rules; returns one error string per violating source.
• TopicsPolicy (site.topics) + indexableTag(tag, postCount, policy) + checkTagPolicy(tags, policy) — mechanical defense against thin tag-page sprawl (v0.3.0). Tag pages that aren't a declared pillar, don't have a cornerstones entry, and sit below indexMinPosts should render noindex,follow; buildSitemap drops them from the sitemap automatically. aliases names known duplicate tags (usa → us) so site validators can warn on them. Omit site.topics entirely for the old index-every-tag behavior.

Each schema is exported as both the Zod value and the inferred TypeScript type.

Client vs server (Next.js)

The @strand-cms/core barrel includes the filesystem loader (node:fs), so it is server-only. Importing it from a "use client" component makes Turbopack try to bundle node:fs for the browser and the app fails at runtime.

• Client-safe: import { postPath, tagPath, authorPath } from "@strand-cms/core/schema" (SiteConfig, RoutesConfig types too — pure zod, no fs).
• Server-only: loadPosts / loadPost / loadAuthors / validatePostFile (from @strand-cms/core or @strand-cms/core/server).
• Pattern: compute link lists in a Server Component (or lib/*.ts server module) and pass plain { href, label } objects into "use client" components. Never import the barrel — or lib/strand.ts if it wraps the loader — from client code.

Loader

Filesystem in, typed content out:

import { loadPosts, loadPost, loadAuthors, isLive } from "@strand-cms/core";

const posts = loadPosts("content/posts");            // live posts, newest first
const post  = loadPost("content/posts", "my-slug");  // one post, drafts included

• loadPosts(dir, opts) / loadPost(dir, slug) — parse + validate MDX files; each Post carries frontmatter, body, and computed reading time.
• loadAuthors(dir) / loadAuthor(dir, id) — author files.
• isLive(frontmatter) — published, or scheduled with a publish time in the past.
• validateFrontmatter(data) / validatePostFile(path) — field-level ValidationResults; these back strand validate and the MCP validate_post tool.

SEO generators

import { buildMetadata, postGraph, sitemapXml, buildRss, buildRobots } from "@strand-cms/core";

• buildMetadata(post, site, routes) — title, description, canonical, per-article keywords, full robots/googlebot preview directives (max-snippet:-1, max-image-preview:large, max-video-preview:-1), ai-content-type / ai-topic hint tags, article:modified_time (from updatedAt, falling back to publishedAt), and OpenGraph + Twitter cards as framework-agnostic PageMetadata. Descriptions are clamped with metaDescription() (Bing SEO/GEO hard window 25–160).
• metaDescription(text) — word-boundary clamp helper for site/tag/dynamic templates.
• articleJsonLd / faqJsonLd / breadcrumbJsonLd / websiteJsonLd / postGraph — JSON-LD (Article/NewsArticle/BlogPosting, FAQPage, breadcrumbs) ready to embed.
• sitemapXml, buildRss, buildRobots — the classic crawl surface.

GEO generators (AI search)

For ChatGPT / Perplexity / Claude / AI Overviews:

• buildLlmsTxt / buildLlmsFullTxt — the llms.txt index and full-content variant.
• mdxToMarkdown(body) — strip MDX to clean markdown (source text, not hydrated DOM).
• renderPostMarkdown(post, …) — the content-negotiated .md version of a post page.

All generators are deterministic over the MDX, so their output belongs in the static build — emit at build time, rebuild on merge.

Used by

@strand-cms/cli (MCP server + validation), @strand-cms/content-api (headless JSON layer), and the default Next.js theme scaffolded by create-strand.

MIT.

Crawl surface (hard rule)

Never platform-redirect /robots.txt / /sitemap.xml / /llms.txt across hosts (Bing Soft 404). Use CRAWL_SURFACE_PATHS, isCrawlSurfacePath, and hostCanonicalRedirectUrl from this package inside Next.js proxy.ts. site.url must be an https origin with no trailing slash.