Your Docs Deserve a Chatbot That Actually Reads Them

You've written the documentation. You've organized it, tagged it, and maybe even added a search bar. But your users are still asking the same questions in Discord, in GitHub issues, and in your support inbox. What if your docs could just answer those questions directly? Docus is a documentation framework that ships with an AI assistant and an MCP server baked right in, so your Markdown files do more than sit there looking pretty.

What It Does

Docus is a documentation framework built on Nuxt 4, Nuxt UI 4, and Tailwind CSS 4. You write your docs in Markdown (with extended MDC components), and it generates a complete documentation site with responsive design, dark mode, client-side search, and SEO optimization including sitemap, robots.txt, and OG image generation.

The interesting part is the AI stack. Every Docus site ships with an embedded chat assistant that answers questions from your documentation, cites its sources, and generates code examples. It's powered by Vercel AI Gateway and your own MCP server. That MCP server is exposed at /mcp, which means you can install it directly into Cursor, VS Code, Claude, or any other AI tool that supports the Model Context Protocol. Your docs become queryable from inside your editor.

Docus also handles the growing ecosystem of AI-readable documentation formats. It automatically generates llms.txt and llms-full.txt files, supports Agent Skills Discovery through a skills/ directory served at /.well-known/skills/ (following the Cloudflare Agent Skills Discovery RFC), and provides internationalization with 20+ locales for the assistant UI.

Why It's Cool

• The MCP server is the killer feature. Most documentation tools treat AI as an afterthought—maybe a plugin, maybe a third-party integration. Docus makes it a first-class citizen. You get a working MCP endpoint at /mcp without writing a single line of server code. That means your users can ask their editor "how do I configure authentication?" and get an answer pulled directly from your docs, with source citations.

• It covers the whole AI documentation surface area. There's the assistant for end users browsing your site. There's the MCP server for developers in their editors. There's llms.txt for the emerging standard of LLM-friendly content. There's even Agent Skills Discovery, which lets you publish reusable skills from your docs. Docus isn't just adding a chat widget and calling it done—it's thinking about how AI tools actually consume documentation.

• The setup is genuinely fast. npx create-docus my-docs and you're running at localhost:3000. No configuration files to wrestle with, no plugin ecosystem to navigate. The CLI scaffolds everything: design, search, dark mode, i18n support, TypeScript. You can go from zero to a deployed docs site in minutes.

• It doesn't sacrifice the basics. Beautiful design, mobile-first responsiveness, dark mode with a keyboard shortcut toggle, full-text search with an optional FTS5 backend—these are things you'd expect from a mature docs framework, and Docus delivers them without making you choose between "nice docs" and "AI-enabled docs."

• Internationalization is built in. Not just for your content, but for the assistant UI itself. If you're writing docs for a global audience, the assistant speaks their language too.

How to Try It

You can have a Docus site running locally in about a minute:

# Create a new project
npx create-docus my-docs

# Or create with the i18n template for multi-language docs
npx create-docus my-docs -t i18n

# Navigate to your project
cd my-docs

# Start the development server
npm run dev

Your site will be available at http://localhost:3000.

If you'd rather start from a deployed template, you can clone the Docus starter directly from Vercel and create your git repository from there. That's a nice option if you want to see the full setup before committing to a local install.

Once you're running, check out the Docus documentation for guides on the assistant, MCP server configuration, and Agent Skills. The repository is at github.com/nuxt-content/docus.

Final Thoughts

Docus makes the most sense if you're already in the Nuxt ecosystem, or if you're starting a new documentation project and want AI features without stitching together five different tools. The MCP server alone is worth the look—it's a clean answer to a problem most docs sites haven't even started thinking about yet. If you're maintaining docs that people actually read (and ask questions about), Docus gives you a way to meet them where they are: in their editor, in their language, with answers pulled straight from your source of truth.

Follow @githubprojects for more developer tools and open source projects.