Wispr Flow: Dictation That Works EverywhereStop typing. Start speaking. 4x faster.
Promoted
briefroom MCP
Your AI agent deploys HTML — humans comment, AI fixes
2 followers
Your AI agent deploys HTML — humans comment, AI fixes
2 followers
An MCP server that lets coding agents (Claude Code, Cursor, Codex…) deploy HTML to a secure share URL and pull reviewer comments back as LLM-ready Markdown. Deploy → share → collect feedback → redeploy, without leaving your agent. On the official MCP registry.
Hi PH! 👋 Happy Vercel Day!
I launched briefroom here a week ago — a service where AI-generated HTML (proposals, mockups, reports) gets a secure share URL and clients can comment directly on the page.
The #1 gap: the human still had to do the sharing. So this week I shipped briefroom MCP.
🤖 What your agent can do now:
• deploy_html — zip & upload a directory, get a share URL back in seconds
• get_feedback — pull client comments as LLM-ready Markdown (paste straight back into your agent)
• list_deployments — manage rooms from inside the conversation
Setup (Claude Code):
claude mcp add briefroom -- npx -y @briefroom/mcp
Auth: set BRIEFROOM_TOKEN (free account) or run `npx @briefroom/cli login` once.
It's on the official MCP registry (io.github.briefroom/mcp), MIT-licensed on GitHub, and wraps the same CLI you can use directly.
That share URL in the first screenshot? It's real — Claude Code deployed that page through this exact MCP server while we prepped this launch. Open it and leave a comment on any element:
https://briefroom.net/s/e3cyjyw6taw4qj6ystcrqkxvol4xkird
I might just feed your comments back to the AI and update the page during launch day. 👀
Fun timing: Claude artifacts just gained MCP superpowers — apps for viewers inside Claude. briefroom covers the other half: sharing results with people outside it. Clients need zero account, and their comments come back as prompts.
Also shipped since launch week:
📄 PDF export with proper Japanese font embedding (server-side subsetting — no font requests leave our infra)
🖱️ Area-select comments (drag a box around anything)
🗂️ Version management — roll back or prune any deploy
briefroom runs on Vercel (Next.js) with an isolated delivery domain on Cloudflare for the user HTML. Fitting day to launch. 🖤▲
Dropping HTML on the site needs no signup for your first deploy; the MCP runs on a free account token. Would love your feedback on the agent workflow. AMA!
@briefroom
日本語でも 🙏
1 週間前に launch した briefroom(AI 生成 HTML に共有 URL を発行し、ページ上で直接コメントをもらえるサービス)の MCP サーバー版です。
エージェントができること:
• deploy_html — ディレクトリを ZIP 化してアップロード、共有 URL を即発行
• get_feedback — クライアントのコメントを LLM 向け Markdown で回収(そのままエージェントに貼り戻せます)
• list_deployments — 会話の中からルーム管理
セットアップ(Claude Code): claude mcp add briefroom -- npx -y @briefroom/mcp
launch 後 1 週間で、PDF エクスポート(日本語フォント埋込対応)/ 範囲を囲ってコメント / バージョン管理(戻し・削除)も追加しています。フィードバック歓迎です!
the review feedback coming back as LLM ready Markdown is genuinely clever, most tools just dump raw comments and call it a day. basically cuts the whole loop down to one agent call which is exactly how it should work.
@turanneamo2woj
Thanks — that's exactly the design goal! The part that makes it work as "one agent call" is that each comment comes back with everything the agent needs to act: the CSS selector, a DOM excerpt, and a screenshot of the exact region the reviewer selected. There's also a header that tells the LLM to treat comment bodies as untrusted data, so a mischievous client can't prompt-inject your coding agent 😄
get_feedback → edit → redeploy to the same URL, and the loop closes. If you're curious about the raw format, drop a comment on the demo page from the first screenshot and pull it — the feedback endpoint is public, no auth needed for reading.