How Docsbook publishes documentation machines can cite

Most companies with a product, customers and an ad budget still cannot be recommended by Google or by an AI assistant, because there is nothing to read: what exists is out of date, hidden behind a login, or disagrees with the product. Docsbook publishes what is already written to where machines read it, and shows how many readers it brought.

What happens between your repository and a published page#

Reach: a page for each question your customers ask#

A documentation page that answers one narrow question costs less to write than a landing page and keeps earning search traffic for years, because the question does not go out of fashion. Docsbook gives each page its own URL, its own title and description, its own entry in sitemap.xml, and its own JSON-LD — so a page about one question competes on that question rather than being buried inside a marketing site.

Docsbook generates the machine-readable surfaces for this automatically: meta tags, sitemap.xml, OpenGraph, canonical URLs, and JSON-LD for FAQ and HowTo content. See SEO for the search-engine surface and GEO for the generative-engine one.

Machine-readable facts: what an assistant quotes about you#

An AI assistant recommending a product quotes what it can read and verify cheaply: prices, limits, what the product does not do, and who the company is. Where your documentation says nothing, the assistant fills the gap with a competitor's page. Docsbook's job here is to make those facts readable — as HTML rendered on the server, as llms.txt, and as an MCP server.

Three surfaces do this work:

• llms.txt — a plain-text index of your documentation at the site root, for AI agents that look for one. See llms.txt.
• MCP server — 140 tools over the Model Context Protocol, so Claude Code, Cursor or ChatGPT can ask the docsbook_expert agent what to do and get instructions back, read your pages, search them, and commit changes back. See MCP server.
• AEO markup — FAQPage, HowTo and speakable JSON-LD generated from your Markdown, for answer boxes and voice assistants. See AEO.

Analytics: which pages are read and where readers stop#

Docsbook records every page view, search query, feedback vote and tracked event on your documentation, and reports them per page. That answers the questions a marketer asks about a docs site: which pages nobody reaches, which searches return nothing, how far a reader gets before leaving, and which countries and languages they arrive from.

Events, goals and funnels let you mark a path through the docs and count how many readers finish it. See Tracking overview for what is measured and Events for the event catalogue.

Who Docsbook is for#

What Docsbook does not do#

Docsbook does not offer SAML SSO, team accounts with role-based access control, a SOC 2 Type II report, or a Data Processing Agreement — all four are on the compliance roadmap, not shipped. There is no WYSIWYG editor aimed at a non-technical author: pages are Markdown files, edited in the web editor, in GitHub, or by an agent over MCP. See MCP security for the current compliance position.

What Docsbook runs on#

Docsbook is a Next.js 16 and React 19 application on Vercel, with Postgres (Neon) and Redis behind it, Drizzle as the ORM, and Paddle as the merchant of record. Markdown is parsed with unified, remark and rehype, highlighted with Shiki, and navigated with markdown-lsp, Docsbook's open-source documentation parser. AI calls are routed through OpenRouter, or through your own provider key.

What it costs#

Docsbook meters four kinds of AI work against each project's own balance:

• Readers (AI Chat) — an AI answer given to a reader of your documentation.
• Admin & AI Agent — an agent run you or a connected agent started.
• AI Translations — translating a page into another language.
• Semantic Index — building the embeddings the AI chat retrieves from.

It also meters machine crawling — the pages AI and search bots read from your published site, past the monthly allowance every plan includes. Hosting, a custom domain and its TLS certificate, people reading, editors, GitHub sync, full-text search, branding and analytics are unmetered. Every new project starts with $1.00 of balance and can claim $5.00 more once it is 3 minutes old; top-ups after that run from $20.00 to $5,000.00. See Pricing for the mechanism and docsbook.io/pricing for current figures.

Next steps#

• Quick start — publish a documentation site, from source to public URL
• Concepts — workspace, project balance, indexing and the other terms used here
• Use cases — the jobs teams hire documentation for
• FAQ — sync, privacy, data ownership and billing questions