- Core Release: Blume 1.0.3 is an MIT-licensed framework that turns Markdown or MDX folders into documentation sites.
- AI Outputs: It produces Astro-based static pages, machine-readable llms.txt files, optional Ask AI, and four read-only Model Context Protocol tools.
- Deployment Boundary: Node.js 22.12 or newer is required, while request-time AI features also need server output and a supported adapter.
- Adoption Test: Teams must decide whether Blume’s generated scaffold and agent access outweigh the operating burden of its interactive features.
Developer Hayden Bleasel has introduced Blume, an MIT-licensed documentation framework that turns Markdown folders into Astro documentation sites for readers and AI tools.
Markdown or MDX, meaning Markdown with embedded JSX-style components, serves as one content set for people and software. Blume builds static pages and exposes raw Markdown at any .md URL, plus llms.txt and llms-full.txt, Copy as Markdown, Open in chat, an optional Ask AI interface, and a hosted Model Context Protocol (MCP) server. MCP lets compatible AI clients call structured tools to search and retrieve documentation.
Blume needs Node.js 22.12 or newer and one Markdown or MDX file. Request-time Ask AI and MCP functions require server output through a Vercel, Netlify, Node, or Cloudflare adapter. Static pages can remain on static hosting, but interactive AI access brings server operations back into the workflow.
How Blume Hides the App Scaffold
Blume’s command-line interface loads blume.config.ts, scans content into a graph, and generates a hidden Astro project under .blume/. Running blume build outputs static HTML to dist/, without requiring users to configure Astro or Tailwind first. During regeneration, Blume rewrites only changed files, keeping authored pages separate from generated application code.
A catch-all route renders pages inside the generated application while the content folder remains the author-facing workspace. Navigation, search, theming, and Open Graph images arrive without separate application setup. Bun, pnpm, npm, and yarn can all run the tool, although Node.js remains required.
Machine access begins with outputs that need no live assistant. Each page can appear through a .md URL, while llms.txt provides a machine-readable index and llms-full.txt consolidates the site’s content. Blume can also combine local files with remote MDX, Notion, or Sanity material, passing those inputs through the same output path.
An optional hosted MCP server adds four read-only tools: search_docs, get_page, list_pages, and get_navigation. Compatible clients can discover the site structure, search its contents, and retrieve a specific page without scraping the visual interface. Because these requests run live, teams choosing MCP or Ask AI must operate a supported server adapter rather than relying on static hosting alone.
Teams needing custom Astro behavior can run blume eject to produce a standalone project. Running the command exposes the generated application for modification while keeping the Blume package as a dependency. Blume hides the initial scaffold without eliminating the runtime or every maintenance decision.
Introducing Blume 🪷
A world-class docs framework for everything you ship. Drop Markdown into a folder and ship a full docs site with no app boilerplate.
→ zero-config setup
→ automatic SEO and AEO
→ 30+ components
→ powered by astro + vite
→ open source, free forever pic.twitter.com/fQSwLd9BM9— Hayden Bleasel (@haydenbleasel) July 13, 2026
Where Blume Fits Among Markdown Doc Tools
Blume enters a category with several active approaches. VitePress converts Markdown into static documentation with Vue-based components and themes, and it publishes llms.txt and llms-full.txt versions of its own documentation. VitePress keeps a stable 1.x line while version 2 remains in preview after an alpha build published on July 6.
Nextra combines Next.js and MDX with file-system routing, search, internationalization, static export, and an Ask AI option. Its version 4.6.1 added Next.js 16 compatibility in December 2025. YAML-configured MkDocs instead organizes Markdown publishing around themes, plugins, live preview, and static deployment.
MkDocs supports themes and plugins, while its version 1.6.1 dates to August 2024, an explicitly historical marker for the established alternative. Blume’s hidden, ejectable Astro runtime keeps application code outside the content folder. Its static files and optional agent tools let teams choose between static hosting and server-backed access.
Agent access to technical documentation predates Blume. Google introduced a developer-knowledge API with documentation access through an MCP server. Blume applies a related interface to documentation that teams keep in their own folders, packaging the reader site and agent surfaces in one open-source project.
Real deployments will test whether teams can run Blume’s Ask AI and MCP features without recreating the application-maintenance work its generated scaffold is designed to reduce. Static-only deployments avoid that server burden, while interactive features require teams to maintain one of the four supported adapters.
