vibedrop — ephemeral hosting for agents via MCP

vibedrop: ephemeral hosting for agents

Connect your agent to an MCP server and publish static HTML to a public<br>URL instantly. No build, no deploy, nothing to configure.

Install MCP

API

$ claude mcp add --transport http vibedrop https://mcp.vibedrop.sh/mcp<br>✓ Added MCP server vibedrop

› create a coffee shop site and upload it<br>→ Write → index.html<br>→ deploy_site → https://k3f9a1b2.vdrop.site

What is vibedrop?

vibedrop is a remote<br>MCP server that gives your<br>agent the ability to publish websites. You pass it text files (HTML, CSS,<br>JS) and it serves them behind a public URL on its own subdomain:<br>your-id.vdrop.site.

There's no build and no deploy pipeline. The agent calls a tool, the files<br>go online instantly, and you share the link. It's ideal for previews, demos<br>and validating ideas fast — everything you publish is<br>ephemeral and auto-deletes<br>after ~24 hours.

Get started

Three steps to publish your first site.

Step 1. Connect the MCP to your agent

Pick your client below. They all point at the same Streamable HTTP endpoint.

Claude Code

OpenCode

Codex

Cursor

Claude Cowork

Other clients

Add vibedrop to Claude Code

Run the command below, then open (or restart) Claude, run<br>/mcp,<br>select vibedrop, and finish the sign-in that opens<br>in your browser. The tools appear once the server shows connected.

$ claude mcp add --transport http vibedrop https://mcp.vibedrop.sh/mcp<br>Copy

Add vibedrop to OpenCode

Add this to opencode.json<br>(or ~/.config/opencode/opencode.json),<br>then complete the browser sign-in when OpenCode prompts you — or trigger it with<br>opencode mcp auth vibedrop.

Copy

"$schema": "https://opencode.ai/config.json",<br>"mcp": {<br>"vibedrop": {<br>"type": "remote",<br>"url": "https://mcp.vibedrop.sh/mcp",<br>"enabled": true

Add vibedrop to Codex

Open Settings → MCP servers → Add new server,<br>pick Streamable HTTP, paste the URL, then complete the browser<br>sign-in Codex opens. The tools are ready once you approve.

https://mcp.vibedrop.sh/mcp<br>Copy

Add vibedrop to Cursor

Add this to ~/.cursor/mcp.json<br>(or .cursor/mcp.json in your project),<br>restart Cursor, then authorize vibedrop in your browser<br>from Settings → MCP.

Copy

"mcpServers": {<br>"vibedrop": {<br>"url": "https://mcp.vibedrop.sh/mcp"

Add vibedrop to Claude Cowork

Open Settings → Connectors → Add custom connector,<br>paste the URL below, then complete the browser sign-in. Because Cowork connects from<br>Anthropic's cloud, only remote endpoints like this one work — no local setup needed.

https://mcp.vibedrop.sh/mcp<br>Copy

Add vibedrop to any MCP client

Any MCP-compatible agent works over the same Streamable HTTP endpoint. Point your client at<br>the URL below and complete the browser sign-in it opens — no API key needed.

https://mcp.vibedrop.sh/mcp<br>Copy

Step 2. Ask it to publish

In plain language. The agent builds the HTML and calls<br>deploy_site<br>for you:

> build a landing page for my coffee shop and publish it

deploy_site(files: [

{ path: "index.html", content: "…" },

{ path: "styles.css", content: "body{…}" },

{ path: "app.js", content: "…" },

]})

Deployed 3 files &middot; index.html, styles.css, app.js

Step 3. Open your URL

Claude Code hands you both links in plain language:

Done — your coffee shop landing page is live:

Live (share this): https://k3f9a1b2.vdrop.site

Dev (live-reloads as you iterate): https://k3f9a1b2-dev.vdrop.site

Heads up: the site auto-deletes in ~24h, and each deploy resets the clock.

The MCP API

The tools your agent can invoke. Every deploy returns<br>url (clean, to share),<br>devUrl (same page with live-reload while you iterate) and<br>expiresAt (when it auto-deletes).

deploy_site

creates / updates

Publishes one or more text files as a static site and returns its public<br>URLs. Pass an existing projectId<br>to update that site, or omit it to create a new one. Updates are incremental:<br>files you leave out stay live.

files required<br>Array of { path, content }. path is the relative path (e.g. index.html or assets/app.js); content is the file's text.

projectId optional<br>Existing id to update. If omitted, a new one is generated.

description optional<br>Short summary so the site is easy to recognize later in list_sites.

replace optional<br>When true, mirror the site to exactly the files you pass, deleting any others. Defaults to incremental.

Per-deploy limits: 100 files · 10 MB per file · 25 MB total.

get_upload_command

fast path

When the agent detects curl<br>is installed, it uses this instead of deploy_site:<br>it returns a ready-to-run curl<br>command that uploads local files straight from disk. Their contents never pass<br>through the agent's context, so it's much faster and cheaper than inlining them,<br>and it supports images and other binaries. Takes<br>paths plus the same<br>optional projectId,<br>description and<br>replace.

list_sites · get_site · get_file

Inspect what you've published. list_sites<br>returns each site as { projectId, url, description, files } (no parameters);<br>get_site returns that same detail...