VibeSharing MCP Server

The missing link between your AI coding assistant and your team.

Note for anyone reading this on GitHub: erova/vibesharing-mcp is a read-only mirror, published automatically on each release. Pull requests against it will be overwritten by the next sync. To report a bug or request a change, please open an issue or write to hello@vibesharing.app.

Build prototypes with Claude Code, Cursor, or any AI tool — then instantly share them with your team for feedback. No context switching. No copy-pasting URLs. Just ship and collaborate.

Quick Start

1. Get Your Token

Sign up at vibesharing.app, then go to Account Settings to:

• Copy your deploy token — starts with vs_

• Connect your GitHub account (optional) — enables Push to Deploy

2. Connect Your Editor

Cursor / Windsurf / Claude Desktop (recommended — always up to date):

{
  "mcpServers": {
    "vibesharing": {
      "url": "https://vibesharing.app/api/mcp",
      "headers": {
        "Authorization": "Bearer vs_YOUR_TOKEN"
      }
    }
  }
}

Add to .cursor/mcp.json (Cursor) or ~/Library/Application Support/Claude/claude_desktop_config.json (Claude Desktop).

Claude Code:

claude mcp add vibesharing -s user \
  -e VIBESHARING_TOKEN=vs_YOUR_TOKEN \
  -- npx -y @vibesharingapp/mcp-server@latest

Then restart your editor. That's it.

3. Start Sharing

You: "Share this page on VibeSharing"
You: "Deploy this to the Product Redesign collection"
You: "What feedback came in on the dashboard?"

Related MCP server: Colab MCP

What You Can Do

Share a static HTML page instantly

"Share this HTML file on VibeSharing"

One command. No Vercel, no GitHub, no React wrapper. Your HTML page goes live with a feedback widget built in. CSS, SVGs, and images are auto-bundled into a self-contained file.

Deploy full projects

"Deploy this to VibeSharing in the Hero Use Cases collection as checkout-v3"

Claude confirms the collection, names the deployment, and handles everything — GitHub repo, deploy, team registration — in one conversation. Your first three prototypes are hosted on us; connect your own Vercel or Netlify account and the limit disappears.

Match your design system from the first prompt

"Build a settings page using our design system"

quick_prototype picks the right template from your org, hands the AI its CSS variables and written design instructions, and deploys — so the prototype is on-brand as it's generated, instead of built generic and retrofitted. Browse what's available with list_templates.

Go through feedback one person at a time

"What feedback did I get on the checkout flow?"
"Walk me through Jordan's comments"

On the local server, open feedback across your recent prototypes is summarized on the first VibeSharing tool call of a session — you find out something is waiting without asking. (Not yet on the remote endpoint.)

Asked without a name, get_feedback groups everything by author — "9 items from 4 people, 6 open" — instead of returning a wall of text. Asked with a name, it walks that person's comments one at a time, each carrying the page it was left on, the pinned position, the question it answered, the replies underneath, and the screenshot itself as an image — so you see what they saw rather than opening a link. You decide per comment whether to build it, skip it, or discuss — and close_feedback_loop tells the person what you did.

Keep every version, and branch safely

"Show me the version history"
"Fork this so I can try a dark variant"
"Roll back to v3"

Every deploy is a numbered version with its own immutable URL. fork_prototype copies one to a separate URL so you can explore without touching the original, and rollback_deploy restores any earlier version — for git-based and static prototypes alike.

Put it in front of customers

"Set up a study comparing these three variants"

create_campaign bundles several prototypes behind one gated portal with your own questions — star ratings, multiple choice, free text — and supports blind randomized A/B/C ordering so a name like "v2" can't skew a preference test. It lands as a draft; opening it and inviting the cohort stays a deliberate step in the dashboard.

Catch build failures before deploying

"Check this will build before we ship it"

validate_project looks for the things that actually break deploys — a missing framework dependency, no build script, conflicting configs — and returns fixes rather than a stack trace after the fact.

Keep context alive across sessions

"Sync my CLAUDE.md to VibeSharing"

Your project context persists on VibeSharing, so any team member (or AI session) can pick up where you left off. CLAUDE.md and AGENTS.md auto-sync on deploy.

Troubleshoot issues

"Run a VibeSharing health check"

The diagnose tool checks your token, GitHub connection, deploy locks, and recent errors — and auto-fixes what it can. If it can't fix something, send_support_request sends a detailed report to the admin.

Available Tools (31)

Remote vs Local

Remote (recommended): Connect to vibesharing.app/api/mcp. Always up to date, no installs, no cache issues. Works everywhere.

Local (advanced): Run via npx. Supports reading files from disk with file_path parameter. Requires Node.js 18+. May need npx clear-npx-cache after updates.

Environment Variables (local only)

Learn More

• VibeSharing — Sign up free

• Setup Guide — Step-by-step for all editors

• Troubleshooting — Common issues and fixes

• GitHub — Source code

Stop building in isolation. Start shipping with your team.