Quickstart

This is the GA happy path: one MDX source, one docs route, and one generated artifact set for agents.

Install

Fastest: scaffold with leadtype init

In an existing app, leadtype init writes the integration for you — a starter docs/ source, the docs route, the markdown route, config, and a generated set of agent artifacts — then runs leadtype generate so the site works on the first dev.

# Auto-detects your framework from package.json, or pass --framework
npx leadtype init --framework next

leadtype init: scaffolded next docs integration
  + docs/docs.config.ts
  + docs/index.mdx
  + next.config.mjs
  + lib/source.ts
  + lib/mdx-components.tsx
  + app/docs/[[...slug]]/page.tsx
  + package.json (added "docs:generate" script)
  + AGENTS.md (created leadtype docs pointer)

leadtype init: generating agent artifacts…

The last line is the root-AGENTS.md pointer: init adds (or merges, never overwrites) a marked block pointing your coding agent at node_modules/leadtype/AGENTS.md, so any agent using leadtype or editing your docs reads leadtype's own version-matched docs. Re-running refreshes the block in place.

Supported targets: next, astro, nuxt, sveltekit. Run with --dry-run to preview the file list, or --base-url, --name, and --summary to seed the config. For TanStack Start and Fumadocs (heavier app-specific setup), follow the recipes in Use the source primitive and Integrate with Fumadocs.

init scaffolds the local-docs/ case. If your docs live in another repo, or you're adding Leadtype to an app with an existing structure, hand a coding agent one of the agent setup prompts instead — it adapts to your real layout.

That's the whole quickstart if you let the CLI scaffold it. The steps below walk through the same wiring by hand — read them to understand what init generated or to customize it. Steps 1, 2, 4, and 5 are framework-neutral; only step 3 (rendering) varies by framework.

1. Author one page

---
title: "My library"
description: "What it does in one sentence."
---

# My library

Welcome.

Every page needs a title. Add description so generated llms.txt, search results, and bundled AGENTS.md can route readers without guessing from the body.

Add a small config next to the docs source:

import { defineDocsConfig } from "leadtype";

export default defineDocsConfig({
  product: {
    name: "My library",
    tagline: "What it does in one sentence.",
  },
  llms: {
    // `llms.sections` is the body of llms.txt, in order. Use markdown blocks for
    // prose/highlights and links blocks for curated entry points.
    sections: [
      {
        type: "links",
        heading: "Best Starting Points",
        links: [
          {
            urlPath: "/docs",
            title: "Start here",
            description: "Overview and first steps for using My library.",
          },
        ],
      },
    ],
  },
  navigation: ["index"],
});

navigation is the shared source of truth for the sidebar, llms.txt, AGENTS.md, sitemap markdown, and Agent Readability metadata.

2. Create the source

import { createDocsSource } from "leadtype";

export const source = await createDocsSource({
  contentDir: "./docs",
  baseUrl: "https://example.com",
});

That's the framework-neutral primitive. You can now call:

• source.loadPage(slug) — resolved markdown + AST + frontmatter + TOC
• source.listPages() — every page's slug, urlPath, and metadata
• source.getNavigation() — grouped sidebar tree
• source.buildSearchIndex() — static search bundle
• source.resolveInclude(specifier) — standalone include resolver

3. Render the docs route

Step 3 is the only step that varies by framework — here's what it looks like in Next App Router. The same createDocsSource() from step 2 powers every framework; pick yours below the code block to see the equivalent wiring.

import { notFound } from "next/navigation";
import { MDXRemote } from "next-mdx-remote-client/rsc";
import { source } from "@/lib/source";
import { mdxComponents } from "@/lib/mdx-components";

export async function generateStaticParams() {
  const pages = await source.listPages();
  return pages.map((page) => ({ slug: page.slug }));
}

export default async function DocsPage({
  params,
}: {
  params: Promise<{ slug?: string[] }>;
}) {
  const page = await source.loadPage((await params).slug ?? []);
  if (!page) {
    notFound();
  }

  return (
    <article>
      <h1>{page.title}</h1>
      {page.description ? <p>{page.description}</p> : null}
      <MDXRemote components={mdxComponents} source={page.markdown} />
    </article>
  );
}

The UI is yours. Leadtype provides page loading, navigation, markdown output, table of contents data, include resolution, and search input data.

Pick your framework

4. Generate agent artifacts

Run the site-mode CLI before your app build:

npx leadtype generate \
  --src . \
  --out public \
  --base-url https://example.com

That writes the hosted files agents need:

Keep this command in your build script so browser docs, markdown mirrors, search, and agent discovery files stay in sync.

5. Verify the output

Before publishing, check the files that prove the pipeline worked:

npx leadtype generate --src . --out public --base-url https://example.com --json
test -s public/llms.txt
test -s public/docs/index.md
test -s public/docs/search-index.json
test -s public/docs/agent-readability.json

Then add runtime handling for markdown responses, sitemap/robots regeneration, and JSON-LD with Serve agent responses.

Common next steps