Moda MCP turns Claude-made HTML into editable, shareable design work.
Anvisha’s post is a product/workflow pitch: Claude Code and Claude chats can generate rich HTML docs, dashboards, decks, and visual artifacts; Moda adds the missing collaboration/editing layer — upload through MCP, edit on a real canvas, comment, password-protect, and share.
What the X post says
@anvisha’s top-level claim
“HTML docs are starting to replace Notion for us. The missing piece was collaboration. So we built a Google Drive for HTML docs with a WYSIWYG editor, commenting, share links, and password protection.”
The post had meaningful traction at capture/fetch time: ~1.1k likes, ~2.2k bookmarks, ~300k impressions.
The concrete workflow
- Install Moda MCP.
- Make an HTML doc or dashboard with Claude.
- Ask Claude to upload it to Moda.
- Edit or request a redesign in Moda.
- Share with a team or externally.
The post quotes Thariq’s X Article “Using Claude Code: The Unreasonable Effectiveness of HTML,” which argues that HTML artifacts are often more readable, dense, visual, and shareable than long Markdown plans.
Moda MCP in one screen
Capabilities pulled from Moda’s product/docs pages
- Slide decks
- Social posts / ads
- Reports and one-pagers
- Editable 2D vector layouts
- Brand kits from a URL
- Colors, fonts, logos
- On-brand visual generation
- Paste a Moda share link
- Generate React/Vue/HTML/CSS
- Extract design tokens
- WYSIWYG editing
- Comments
- Share links
- Password protection
- Claude.ai
- Claude Desktop
- Claude Code
- Codex, Cursor, VS Code
- Hosted server
- OAuth auth flow
- No local server required
https://mcp.moda.app/mcp
Setup notes if Ananth wants to test it
Moda’s docs describe a hosted MCP server rather than a local install. For Claude Code, the documented command is:
claude mcp add --transport http moda https://mcp.moda.app/mcp
Then run /mcp inside Claude Code and authenticate the Moda server in the browser. For claude.ai / Claude Desktop, add a custom connector named Moda with the same MCP URL, then authenticate and enable it for the chat.
This explainer does not execute setup. It only captures the workflow and fit.
Connection to the “HTML effectiveness” thesis
Thariq’s quoted article gives the underlying reason this was interesting enough to save: Markdown is convenient, but agent work is increasingly too visual, long, and decision-heavy for plain text. HTML can carry layout, SVG diagrams, tables, interactive controls, annotated diffs, and responsive reading structure.
- Exploration grids and design alternatives
- Implementation plans with mockups and flow diagrams
- PR / code explainers with annotated diffs
- Research reports and learning references
- Throwaway editors that export JSON/Markdown/prompt text
- HTML takes longer to generate than Markdown.
- Diffs are noisy under version control.
- Quality depends on taste/design instructions.
- For private canonical state, source data should still live in structured files — HTML is the readable artifact.
Dab’s read: where Moda fits
If an artifact needs to be shown to someone else, iterated visually, or made into a deck/social/report, Moda could be the handoff layer after Claude creates the first draft.
Mission Control should not depend on Moda as source of truth. Use Moda as an optional publishing/editing surface, not as the canonical capture database.
Claude writes items/<slug>/explainer.html or projects/*.html → when shareable/polished output is needed, Claude uploads to Moda → human edits → final share link stored as a structured artifact.
Does Moda preserve enough structure and exportability to round-trip back into local files, or is it mostly a one-way publishing/collaboration canvas?
Sources inspected
- Anvisha X post and referenced parent tweet, fetched via X API/xurl.
- Moda MCP landing page, including metadata and visible setup copy.
- Moda MCP getting started docs, fetched directly from the public docs page.
- Thariq X Article, fetched via X API
tweet.fields=article.
Limitations: no authenticated Moda account was used; no MCP connection or upload flow was executed; the X video itself was not transcribed, so this is based on the tweet text, linked pages, and X Article body.