Leadtype

Leadtype is a docs pipeline for teams that want one MDX source to serve humans and agents. Write docs once, render them in your app, then generate the files agents need: markdown mirrors, llms.txt, root llms-full.txt, search JSON, Agent Readability metadata, and package-bundled AGENTS.md output.

It is not a docs website framework. Bring your own host and UI — a custom Next.js app, TanStack Start, Nuxt, Astro, SvelteKit, Fumadocs, anything — and let leadtype handle conversion, validation, search, and the agent-facing outputs it specializes in.

Rendering diagram...

Pick your path

You're building a docs website — Quickstart — render MDX in your app and generate llms.txt, markdown mirrors, search, and discovery files for HTTP agents.

You're shipping docs inside an npm package — Bundle docs into a package — generate an AGENTS.md plus per-topic markdown that ships in the tarball so coding agents read version-matched docs from node_modules. Agents that install your package then run cheaper and stop confidently guessing wrong about your API — see the evals.

You have a site without a docs tree — a CMS-backed blog, a marketing site, a data-driven app — Generate artifacts without a docs tree — hand generateAgentArtifacts() an in-memory page list and get the same agent artifact set, no .mdx files required.

Just want the mental model first? — How it works — the pipeline, the two output modes, and every artifact named once.

Key terms

These show up throughout the docs. The full glossary lives in How it works → Vocabulary.

Site mode — the default leadtype generate. Writes website artifacts to public/: llms.txt, markdown mirrors, search index, sitemap, robots, Agent Readability files.
Bundle mode — leadtype generate --bundle. Writes AGENTS.md + relative docs/*.md into a package directory so coding agents can read them from node_modules after install. If agents.mcp.enabled is set, it also includes package-local MCP retrieval files.
Agent Readability — the structured discovery and attribution data (agent-readability.json + JSON-LD + sitemap + markdown alternate links) that lets agents fetch, identify, and cite docs without scraping HTML.

What leadtype can do

Docs pipeline

One MDX source — local folders, mounted collections, or remote git repos pinned to a ref — converted, validated, localized, and rendered in your app via createDocsSource() without leadtype owning your UI.

AEO & Agent Readability

llms.txt, markdown mirrors with content negotiation, JSON-LD identity, robots Content-Signals, sitemaps, skills, and a docs MCP server — the full agent surface, mapped against the public scoring rubrics.

Agent artifacts without a docs tree

generateAgentArtifacts() emits the same agent surface from an in-memory page list — for CMS blogs, marketing sites, data-driven pages, and microfrontend fragments.

Writing for agents

Frontmatter contracts, agent-safe MDX components, and the content rule that moved the evals: document the non-obvious, not your type signatures.

Search & AI answers

A static search index, framework hooks, source-grounded AI answers, and search tools agents can call via WebMCP.

Version-matched package docs

Bundle AGENTS.md and markdown inside npm packages so coding agents read node_modules/<pkg>/AGENTS.md — they run cheaper and stop guessing wrong about your API.

Leadtype is a pipeline, not a website framework — it pairs with a custom docs app or layers under Fumadocs, Starlight, and friends. See Methodology for where it fits.