How do I scrape JSON-LD with MCP?

Scrape JSON-LD with MCP by calling AgentFetch's fetch_url tool with format="json" — it reads <script type="application/ld+json"> blocks along with OpenGraph and meta tags while extracting the page, and returns the resolved fields in the response metadata. Example: fetch_url("https://example.com/recipe", format="json") returns the page as structured JSON rather than markdown. JSON-LD is the most reliable structured-data source on the modern web — ~40% of crawlable pages publish it, including most major news (NYT, WaPo, Reuters), recipe sites (AllRecipes, Serious Eats), e-commerce (Shopify-default themes), and event listings. When JSON-LD is present, extraction is essentially free of model tokens — the data is already structured, so AgentFetch returns 100-500 bytes of clean JSON vs sending 30KB of HTML through an extraction LLM. For pages without JSON-LD, AgentFetch falls back to OpenGraph + Twitter Cards meta tags (covers another ~30% of pages), then to a small extraction model on clean markdown (handles the long tail). This fallback chain is invisible to the calling agent — it just gets clean JSON back. Avoid manually parsing JSON-LD client-side; pages often ship multiple blocks (Product + BreadcrumbList + Organization), and AgentFetch resolves them into one clean metadata object for you.