Building an AI-ready website with llms.txt, JSON-LD, and MCP
LyvecaAI describes a four-layer technical stack — `llms.txt`, JSON-LD structured data, explicit AI crawler permissions, and an MCP server — designed to make a website visible and callable by AI agents.
Score breakdown
The post demonstrates a concrete, end-to-end implementation of MCP server tooling alongside `llms.txt` and structured data on a production website, illustrating how the agentic web stack can be assembled today with existing open standards.
- 01A `/llms.txt` plain-text file serves as a README for AI agents; estimated 30 minutes to implement with no library needed.
- 02JSON-LD structured data covers Organization, Person, WebSite, and FAQPage (8 Q&A pairs), with bidirectional founder/worksFor links and sameAs references to X and GitHub.
- 03Claude Code generated most of the JSON-LD from a business description; total time was 2–3 hours.
LyvecaAI's post on Dev.to outlines a four-layer technical approach to making a website legible and actionable for AI agents, framing the gap between web designers and AI developers as the core problem most sites currently face.
A plain-text `/llms.txt` file — described as a "README for AI agents" — takes about 30 minutes to add with no library required.
The first two layers are relatively low-effort. A plain-text `/llms.txt` file — described as a "README for AI agents" — takes about 30 minutes to add with no library required. JSON-LD structured data covers `Organization`, `Person`, `WebSite`, and `FAQPage` schemas, with a bidirectional `founder`/`worksFor` link between the organization and founder entities, plus `sameAs` references to X and GitHub. The author notes that Claude Code generated most of the JSON-LD from a business description, with manual review and refinement, taking 2–3 hours total. A third layer — explicitly listing AI bot names (`GPTBot`, `ClaudeBot`, `PerplexityBot`, `GoogleExtendedBot`, `Applebot-Extended`, `cohere-ai`) in `robots.ts` — is described as zero technical complexity but commonly overlooked, since default `robots.txt` configurations often leave these crawlers unspecified or blocked.
The fourth and most complex layer is an MCP (Model Context Protocol) server at `https://lyveca.com/api/mcp`, exposing three tools: `get_pricing`, `get_service_info(topic)`, and `book_meeting(name?, email?)`. It is built with `@modelcontextprotocol/sdk` on Next.js API routes, deployed as a stateless Vercel serverless function, and discoverable via a `/.well-known/mcp.json` manifest. The author estimates 2–3 days for a first implementation, crediting Claude Code for boilerplate generation. As of the post, Google indexed the site within 48 hours of sitemap submission; AI citation visibility is described as still pending external corroboration, with a follow-up planned after retesting in Perplexity and ChatGPT Browse.
Key facts
- 01A `/llms.txt` plain-text file serves as a README for AI agents; estimated 30 minutes to implement with no library needed.
- 02JSON-LD structured data covers Organization, Person, WebSite, and FAQPage (8 Q&A pairs), with bidirectional founder/worksFor links and sameAs references to X and GitHub.
- 03Claude Code generated most of the JSON-LD from a business description; total time was 2–3 hours.
- 04A `robots.ts` file explicitly allows GPTBot, ClaudeBot, PerplexityBot, GoogleExtendedBot, Applebot-Extended, and cohere-ai.
- 05An MCP server at `https://lyveca.com/api/mcp` exposes three tools: `get_pricing`, `get_service_info(topic)`, and `book_meeting(name?, email?)`.
- 06The MCP server is built with `@modelcontextprotocol/sdk` on Next.js API routes, deployed stateless on Vercel, and discoverable via `/.well-known/mcp.json`.
- 07Google indexed the site within 48 hours of sitemap submission; AI citation results are still pending external corroboration.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 10, 2026 · 15:34 UTC. How this works →