Leadtype
Changelog

Leadtype 0.2

Last updated June 3, 2026

Released June 3, 2026.

Leadtype 0.2 turns the package from an MDX-to-agent-artifacts pipeline into a portable docs and agent-readability layer for your own docs host. Use it under an existing stack like Fumadocs, Starlight, or Mintlify-adjacent workflows, or build a custom docs app around the source primitive. The release adds discoverable agent surfaces, a docs MCP server, config-owned navigation, remote source sync, i18n, scoring, and framework examples that show how to keep rendered docs UI separate from the pipeline.

Agent surfaces

  • Added leadtype mcp and leadtype/mcp so generated docs can be served to MCP clients over stdio or mounted as a stateless Streamable HTTP handler. The default tools are search-docs and get-page, with list-pages available opt-in.
  • Added leadtype mcp --check for a one-shot local verification before wiring an editor or hosted MCP client.
  • Added leadtype generate --bundle --mcp so an installed dependency can expose version-matched bundled docs through leadtype mcp --package <name>.
  • Added /.well-known/agent-skills for hosted docs and bundled SKILL.md output for package docs. The default docs skill points agents at llms.txt, MCP, or bundled AGENTS.md depending on the surface.
  • Expanded /.well-known/agent-card.json into an A2A-style AgentCard with provider, documentation URL, skills, capabilities, modes, and the MCP endpoint as the standard url when MCP is enabled.
  • Added /.well-known/llms.txt, /.well-known/llms-full.txt, Link and X-Llms-Txt discovery headers, robots/content-signal generation, SEO metadata, and config-driven JSON-LD.
  • Added leadtype score and leadtype/score to rate a generated build's leadtype-addressable agent readiness and gate CI with --min.

Configuration and source model

  • Added top-level product, organization, agents, and llms.sections config fields so llms.txt, AGENTS.md, JSON-LD, robots, the agent card, and generated skills share one identity model.
  • Added config-owned navigation with nested sections, explicit pages, wildcard includes, root-relative entries, per-node bases, and optional: true sections. Frontmatter group still works as taxonomy and fallback metadata, but curated IA should now live in config.
  • Added collections and leadtype sync for multi-source docs. Collections can read local folders or remote git repositories pinned to a branch, tag, or commit, then generate under separate URL prefixes.
  • Added sourceConfig: true for the flagship pinned source docs UI path. A docs UI repo can pin a source/package repo, inherit source-owned navigation, legacy groups, frontmatterSchema, flatteners, and mounts, then keep product identity, agent policy, deployment, and the app shell in the UI repo.
  • Added first-class docs i18n helpers and locale-aware generation for localized sources, search indexes, LLM files, and Agent Readability artifacts.
  • Added frontmatter transformers and tightened the default frontmatter metadata contract, including stronger linting around JSON-LD validity, link structure, and GEO-oriented docs shape.
  • Added resolveAgentInputs(config) so custom scripts can map config identity into lower-level generation APIs the same way the CLI does.

Host and framework integration

  • Formalized the package boundary: leadtype ships data, types, adapters, handlers, and CLI output, but no rendered DOM. Docs apps own their UI components, layouts, accessibility, and styling.
  • Added and dogfooded framework surfaces and examples for Next.js, Nuxt, Astro, SvelteKit, TanStack Start, and Fumadocs.
  • Added runtime helpers for markdown response negotiation, sitemap/robots responses, docs head metadata, JSON-LD rendering, markdown mirror routing, and Agent Readability manifests.
  • Added search adapters and client helpers for React, Vue, Svelte, Node, Vercel, TanStack, Cloudflare, AI SDK flows, and bash-tool flows.
  • Added authoring docs and component flattening coverage for agent-readable output, including Prompt, FileTree, CommandTabs, and custom flatteners.

Authoring and bundle improvements

  • Added defineComponentFlattener for custom MDX components that need a first-class markdown representation.
  • Improved type table source-root resolution and source-preset behavior so generated tables resolve against the intended project root.
  • Reframed package docs bundling around the most reliable wins from evals: lower token cost and fewer confidently wrong answers.
  • Made leadtype init write or refresh a root AGENTS.md pointer to node_modules/leadtype/AGENTS.md, so agents setting up docs read version-matched Leadtype guidance first.
  • Improved missing-artifact errors for MCP and scoring by listing the actionable fixes: run leadtype generate, point --artifacts at the generated output, or use --package for bundled docs.

Migration notes

  • Move curated sidebars and top-level docs areas into navigation. Keep frontmatter group for taxonomy or compatibility only, and keep order only where wildcard includes need fallback sorting.
  • Replace older ad hoc llms.txt body setup with llms.sections. Use markdown blocks for overview/context and links blocks for best starting points.
  • Add product and organization metadata before enabling agent surfaces so JSON-LD, AgentCard, skills, and generated docs all describe the same entity.
  • Configure agents.robots, agents.seo, agents.mcp, and agents.skills in config instead of duplicating those settings in routes or scripts.
  • For package docs, continue using leadtype generate --bundle. Add --mcp only when the package should ship the search index and Agent Readability manifest needed for package-local MCP retrieval.
  • For docs spread across repositories, use a root leadtype.config.ts with collections, pin remote sources to a reviewed ref, set sourceConfig: true when the source repo owns navigation/schema/flatteners/mounts, and run leadtype generate --sync.
  • For host integrations you control, wire the runtime route to generated agent-readability.json, add markdown response negotiation, render JSON-LD from the manifest, and keep visual components in the host app or docs framework.

Upgrade prompt

Upgrade a docs host integration to Leadtype 0.2

Upgrade this project to Leadtype 0.2 and adopt the new docs-source, generated-artifact, and host-integration best practices. Keep Leadtype as the content and agent-readability layer. If the project already uses a docs framework such as Fumadocs, Starlight, Mintlify, or a custom app, integrate with that host instead of replacing it unless the user explicitly asks for a rebuild.

Inspect first:

  • The installed Leadtype version, package manager, framework, docs source folders, and current generate scripts.
  • Existing docs.config.ts, leadtype.config.ts, frontmatter group and order usage, generated navigation files, and host sidebar/header integration code.
  • Whether the docs are local, split across folders, pulled from remote repositories, or bundled into an npm package.
  • Whether the project should expose agent retrieval through hosted MCP, editor stdio MCP, package-bundled MCP, or no MCP.

Implement:

  • Upgrade Leadtype and update scripts to run leadtype generate with the correct --src, --out, --base-url, and source mode.
  • Define product, organization, llms.sections, and config-owned navigation. Use explicit pages for primary flows and wildcard includes for inventory-like folders.
  • Keep frontmatter group only as taxonomy or compatibility metadata. Keep order only when wildcard navigation includes need fallback sorting.
  • If docs come from multiple local or remote sources, create a root leadtype.config.ts with collections, pin remote refs, set sourceConfig: true when a source repo owns its docs navigation/schema/flatteners/mounts, and use --sync, --refresh, or --offline intentionally.
  • Configure agents.robots, agents.seo, agents.skills, and agents.mcp where relevant.
  • Update only the host integration you control to load the generated Agent Readability manifest, serve markdown responses for agent requests and .md URLs, render JSON-LD/head metadata from Leadtype helpers, and keep all visual UI components owned by the app or docs framework.
  • Add or refresh runtime MDX components for the Leadtype component contract, including Prompt, CommandTabs, FileTree, Tabs, Steps, Callout, and TypeTable.
  • If serving MCP is in scope, install @modelcontextprotocol/sdk in the app that runs the server, validate the generated artifacts before configuring clients, then wire either editor stdio MCP or a hosted route with createMcpHandler.
  • If bundling package docs, run bundle generation for the package. Add MCP bundle artifacts only when consumers should be able to query the installed package's docs locally.

Validate:

  • Run docs generation, linting, type checks, and the app's relevant tests.
  • Confirm /llms.txt, /.well-known/llms.txt, /llms-full.txt, markdown mirrors, search files, agent-readability.json, sitemap, robots, AgentCard, and agent-skills output exist when configured.
  • Confirm llms.txt, generated markdown, Agent Readability navigation, and the human sidebar agree.
  • Confirm markdown responses return Content-Type: text/markdown for agent requests and .md URLs.
  • If MCP is enabled, confirm the configured MCP surface can search and fetch a page before configuring clients.
  • Summarize changed config, routes, scripts, generated artifacts, and any remaining manual docs IA decisions.
Upgrade docs source to Leadtype 0.2

Upgrade only the docs source and Leadtype configuration to the 0.2 best practices. Do not edit app routes, framework layouts, rendered UI components, middleware, or hosting code unless a broken docs-source assumption makes that unavoidable.

Inspect first:

  • docs.config.ts, leadtype.config.ts, collection configs, and generate scripts.
  • The full docs tree, including current frontmatter title, description, group, order, status, and related fields.
  • Existing sidebar/nav conventions in config or generated nav files, but do not change runtime UI code.
  • The highest-value reader and agent entry points: overview, quickstart, install/setup, migration, troubleshooting, API/CLI reference, and changelog.
  • Current leadtype lint and leadtype score output if generated artifacts already exist.

Implement:

  • Move intentional docs IA into config-owned navigation. Use nested sections for the product's real information architecture, explicit page entries for primary flows, and wildcard includes for inventory-like folders.
  • Keep frontmatter group only as taxonomy or compatibility metadata. Keep order only where wildcard includes need fallback sorting.
  • Add or refresh product, organization, and llms.sections. Include an Overview markdown section and a Best Starting Points links section with concise descriptions that route agents to the right pages.
  • If the source spans multiple folders or repos, model it with collections, defineCollection, stable prefix values, pinned remote ref values, sourceConfig: true for source-owned docs config, and clear include/exclude filters.
  • Review every important page against the Write for agents & GEO guidance: lead with the answer, phrase H2/H3 headings as user questions where natural, make sections self-contained, use one term per concept, include exact defaults/limits/failure modes, and avoid restating types or CLI help.
  • Fix mechanical GEO issues: no skipped heading levels, every fenced code block has a language, and every meaningful image has alt text.
  • Add or improve frontmatter descriptions, related links, and page titles so llms.txt, search snippets, sitemap metadata, and Agent Readability output have useful summaries.
  • Preserve public URLs unless a URL is already broken. If a move is unavoidable, document the old and new paths for the UI/hosting owner to redirect.

Validate:

  • Run leadtype lint on the docs source and fix relevant warnings, especially geo:*, broken links, missing descriptions, and unflattened components.
  • Run leadtype generate with the project's normal source/output flags.
  • Run leadtype score --src <docs-source> --out <generated-output> when generated site artifacts exist, and list the remaining score fixes.
  • Confirm every explicit navigation page resolves, wildcard sections do not pull unrelated pages, and llms.txt shows the intended sections and best starting points.
  • Summarize the final navigation tree, the pages changed for GEO/content quality, and any runtime/UI follow-up that should be handled separately.