Build your docs

AI resources

Use the starter's Markdown exports, MCP endpoint, context actions, and optional voice assistant with AI tools.

Updated May 11, 2026

This page collects the AI-native surfaces in the starter. Use them when you want AI tools to read the docs in a structured, source-aware way.

Markdown resources

The starter exposes Markdown for humans and tools:

  • /llms.txt: A compact index of listed pages.
  • /llms-full.txt: A full Markdown export of listed pages.
  • /<slug>.md: A per-page Markdown export.
  • /api/markdown/<slug>.md: The direct Markdown API route.

Use View as Markdown from the page action menu to inspect the generated Markdown for the current page.

MCP resources

The /mcp route exposes tools, resources, and prompts for MCP clients.

MCP capabilities
PropertyTypeDescription
search_docsRequiredtool

Search public page and section content.

read_docRequiredtool

Read a page as Markdown by slug.

list_api_endpointsOptionaltool

List generated OpenAPI and AsyncAPI endpoint pages.

docs://llms-full.txtOptionalresource

Read the full Markdown export through MCP.

Copy the MCP server URL from the page action menu, then add it to your MCP-compatible client.

Assistant citations

The optional text assistant receives the public docs corpus, uses local search results as citation sources, and shows source chips below grounded answers. Add OPENAI_API_KEY only when you want chat. Search, Markdown exports, and MCP work without an AI provider key.

Voice assistant

The voice assistant is disabled by default. Enable it only when you want a browser-based Audio panel with voice presets and live transcript history.

Do

Set NEXT_PUBLIC_DOCS_VOICE_ASSISTANT=true, DOCS_VOICE_ASSISTANT_ENABLED=true, and keep OPENAI_API_KEY on the server.

Don't

Put a standard OpenAI API key in browser code or a public environment variable.

When enabled, /api/voice/session creates an ephemeral OpenAI Realtime client secret. The browser uses that short-lived secret to start a WebRTC session. Set OPENAI_REALTIME_VOICE_FEMALE, OPENAI_REALTIME_VOICE_MALE, and OPENAI_REALTIME_TRANSCRIPTION_MODEL when you want to customize the presets and transcript model.

AI context actions

The page action menu can copy page context or open a provider prompt. Configure visible actions in site.config.ts through siteConfig.contextual.options.

Next step: Search, AI, and MCP

Was this page helpful?