Most of what developers publish isn't code: it's slide decks, product docs, PDFs with no bookmarks, blog posts and architecture diagrams. The friction is usually in the last mile, where plain text has to become something a reader can open. I keep short notes on the third-party tools that remove that friction. These seven are the ones I come back to; the opinions are mine, the projects belong to their authors.
Slides as a Markdown file
If a deck can live in git, it can be reviewed, diffed and rebuilt. Slidev takes that literally: one slides.md, extended Markdown, real code blocks, Mermaid diagrams, and per-slide HTML/CSS when the default look isn't enough. It exports to PDF or PNG, or builds into a single-page app you can host statically. The write-up I keep is about deploying to Cloudflare Pages, including the Playwright detail needed for a downloadable PDF. Details in my Slidev notes for people who dislike PowerPoint.
Docs that are meant to be read by agents too
Mintlify pitches its hosted docs as infrastructure for both people and AI agents, with self-updating content, access control and connections to existing systems. The page leans on adoption numbers and customer case studies, so read it as a vendor's claims rather than a benchmark. Still, the framing — docs as something an agent queries, not just a site humans skim — is worth noticing. Here's what Mintlify says about itself.
One Markdown source, several Chinese platforms
Wenyan is a Markdown formatter that renders for WeChat Official Accounts, Zhihu and Toutiao from the same source. It ships as a desktop app, a web app, a CLI for CI pipelines and an MCP server, which makes it a rare case where the same renderer serves a hand-written post and an automated pipeline. I summarised the whole family in this overview of the Wenyan toolchain.
Letting an AI client push a draft to WeChat
The MCP flavour is the interesting one if you already work through Claude Desktop. It converts Markdown to WeChat rich text, uploads local images and the cover, and drops the result in the draft box, with theme changes by natural-language request. Two practical traps: you must set the app ID and secret as environment variables, and the machine's IP has to be on the account's allow-list or uploads fail. The setup is covered in the wenyan-mcp note.
Putting a table of contents back into a PDF
Plenty of PDFs, especially ones exported from TeX, have no outline, which makes a 400-page book painful to navigate. pdf.tocgen infers headings from font attributes and position: pdfxmeta extracts the heading metadata into a recipe, pdftocgen builds the ToC from it, and pdftocio writes it into the file. It won't help with scanned documents, which is the main limitation to know. The pdf.tocgen workflow is written up in my garden.
Diagrams from a sentence, still editable
smart-excalidraw-next turns a natural-language description into an Excalidraw diagram: flowcharts, architecture, sequence and ER diagrams, mind maps, 20+ types in all. Because the output is real Excalidraw, you tweak it on the canvas instead of re-prompting. You can bring your own OpenAI or Anthropic key, or an admin can configure a shared server-side model behind a password. It runs on Next.js and starts with pnpm dev. See the smart-excalidraw-next summary.
A blog with GitHub Issues as the database
The most frugal publishing stack in this list: each post is an issue, comments come from Gitalk, images live on a separate branch served by jsDelivr, and Vercel deploys the front end. Reading posts through GitHub's GraphQL API even gives you rendered HTML, so you skip the Markdown pipeline. The one rule is to keep the token out of the repo. The full recipe is in this GitHub-as-database blog guide.
What these have in common is that none of them asks you to leave plain text behind: slides, docs, WeChat posts, PDF outlines, diagrams and even a whole blog stay as files you can version. Pick the one that matches the friction you hit this week.
Top comments (0)