The Content Architecture
A production Sanity starter kit built over six years of client work, made for agentic development. Two editions, Next.js and Astro, on one content architecture. AGENTS.md, scoped skills, and MCP servers mean Claude Code and Cursor build inside committed decisions instead of reinventing them.
The Sanity setup agents don't reinvent.
Every Sanity project starts the same way. Spin up Next or Astro, wire the Studio, rebuild the page builder, re-wire draft mode and revalidation. Three days gone. Hand the setup to an agent and you get a different architecture every run.
The Content Architecture is the production foundation underneath my client work, extracted and committed. Hundreds of decisions about how content is modeled, fetched, and composed, made once over six years. It ships in two editions, Next.js and Astro, on the same schema and the same content model. Clone one, rename it, ship.
It's built for agentic development on both ends. AGENTS.md and a dozen scoped skills mean Claude Code or Cursor ingests the conventions before your first prompt, and two MCP servers let the agent read the running app and verify its work in a real Chrome. The shipped site serves an editable llms.txt and a Markdown version of every page to agents that read it.
Underneath: schema as a system, the hard link and media fields already built, a fetch layer where stale-after-publish is solved, a Studio editors navigate without emailing you, SEO at the schema level. Next.js 16 or Astro, Sanity v6, TypeScript strict, Tailwind 4, Biome. The architecture is fixed; the tools are swappable defaults.