A Markdown publishing system for humans and AI agents.
RenderMark turns Markdown into polished documents, hosted pages, PDFs, Word files, HTML exports, slide decks, visual diffs, and agent-ready publishing workflows. The project started as a lightweight browser tool, then expanded into a broader document infrastructure layer: rendering core, web app, VS Code extension, MCP servers, CLI, and agent-integration package.
Markdown is an excellent working format, especially for developers and AI agents, but it often breaks down at the moment of sharing.
A README, proposal, launch plan, or product spec may be structurally complete while still looking unfinished to a client, teammate, stakeholder, or hiring manager.
Existing workflows tend to force a tradeoff: keep the clean source file and accept rough presentation, or move the work into a heavy document editor and lose the simplicity that made Markdown useful.
The document source should stay boring.
The publishing path is where the polish belongs: rendering, typography, export options, share links, access controls, sync, and agent operations should wrap around Markdown instead of replacing it.
That framing made RenderMark less like a text editor and more like a distribution layer for structured writing.
RenderMark is organized as a product system with several connected layers:
The web app supports the normal document lifecycle: create or import Markdown, edit with preview, publish a shareable page, export to PDF/DOCX/HTML, manage public/private access, sync from GitHub, group documents into collections, and manage API keys.
The backend uses Prisma and Neon for users, documents, revisions, API keys, shares, collections, custom domains, OAuth records, and subscription state. Stripe handles checkout, subscription management, and billing webhooks. NextAuth provides Google and GitHub sign-in.
Public documents get dynamic OpenGraph images, optional bylines, indexing controls, password or email-restricted access, and embeddable views.
RenderMark became more interesting once it stopped being only a web app and started acting like document infrastructure for AI agents.
The local MCP server exposes 17 tools for rendering, validation, publish/share workflows, reads and updates, deletion with confirmation, GitHub sync, PDF/DOCX/HTML export, batch export, image output, visual diffs, Google Docs publishing, API-key setup, and workflow guidance.
The hosted MCP server exposes a smaller 10-tool surface for remote clients where filesystem access would be unsafe or impossible. The remote endpoint uses OAuth and per-user credentials, includes MCP safety annotations, and keeps local-only capabilities out of hosted schemas.
The project has multiple entry points because the user might be a person in a browser, a developer in VS Code, or an agent operating inside a coding surface.
The app lives at rendermark.app. The rendering and MCP packages are distributed through npm. The VS Code extension is available on the Marketplace. The public `rendermark-agents` repo documents setup paths for Codex, ChatGPT, Claude, Cursor, Gemini, Copilot, Windsurf, and standards-based MCP clients.
That repository also includes validation automation for the standalone skill, plugin wrapper, marketplace wiring, credential placeholders, and hosted MCP/OAuth discovery contract.
RenderMark now works as a cohesive publishing layer rather than a single-purpose converter.
It gives Markdown users a path from source text to shareable, styled output, while giving AI agents explicit tools for rendering, validation, publishing, syncing, and document management.
The most useful shift was designing the system around both human UX and agent UX. The same product had to feel simple in a browser, predictable through an API, clear inside VS Code, and safe enough for autonomous tools.
The product is a useful example of designing at the boundary between AI tooling and traditional SaaS.
Agents are very good at producing Markdown, but the last mile still matters: trust, typography, export fidelity, access control, versioning, credentials, and clear permission boundaries.
RenderMark makes that last mile explicit.