VibeKit MCP Server
A Model Context Protocol server that lets Claude Desktop, Cursor, Zed, Continue, any MCP client, operate your VibeKit account directly. Deploy GitHub repos, set env vars, tail logs, chat with hosted agents. Tell Claude "deploy this and set the API key"; Claude does it.
Remote connector, no install
For clients that take a remote MCP URL (claude.ai web connectors, ChatGPT, and more), there's nothing to install, just paste this URL:
https://mcp.vibekit.bot/mcp?api_key=vk_your_api_key_hereClients that let you set headers can use the bare https://mcp.vibekit.bot/mcp with Authorization: Bearer vk_your_api_key_here. Get a key from @the_vibe_kit_bot on Telegram with /apikey.
Already added a vibekit server, or rotating your key? Prefer the header form. Because the ?api_key= URL bakes the key into the endpoint, changing keys means replacing the whole server entry, and some clients keep the old one until you explicitly overwrite it, so you can end up authenticating with a stale key without noticing. With the bare URL + Authorization: Bearer header (or the VIBEKIT_API_KEY env var for local/stdio), the endpoint never changes, so rotating a key is just swapping the header value. If your client added the URL form, delete the existing vibekit entry before re-adding rather than adding a second one.
Install (local / stdio)
Prefer running it locally (e.g. Claude Desktop, Cursor)?
npm install -g vibekit-mcpGet a VibeKit API key from @the_vibe_kit_bot on Telegram with /apikey.
Configure Claude Desktop
Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or the equivalent on Windows / Linux:
{
"mcpServers": {
"vibekit": {
"command": "vibekit-mcp",
"env": {
"VIBEKIT_API_KEY": "vk_live_..."
}
}
}
}Restart Claude Desktop. The new tools appear under the connectors icon.
Configure Cursor
Cursor → Settings → Features → Model Context Protocol → Add a new MCP server. Same command + env vars.
What you can do
36 tools exposed today, grouped by surface:
Hosting & Apps
AI Agent
Database
QA
Tasks
Schedules
Account
Skills
Why this beats the dashboard
- Stay in your editor. No tab-switching. Deploy from inside the same chat you're writing the code in.
- Compositional. Claude can chain calls, "deploy this, wait for it to come up, then set these 3 env vars, then restart", without you nursing each step.
- Scriptable from agents. If you have a Claude Code workflow that needs to deploy a sibling app, MCP makes it one tool call.
Related packages
- vibekit-agent, run on your laptop to be a remotely-controllable Claude Code worker (iOS Remote Agent path)
- vibekit-cli, terminal CLI for kicking off agent tasks and scheduling cron jobs
FAQ
What does the VibeKit MCP server do?
It exposes your VibeKit account as tools that any MCP client (Claude Desktop, Cursor, Zed, Continue, etc.) can call. You tell Claude Desktop 'deploy this repo and set OPENAI_API_KEY' and Claude actually does it via the MCP tools, no copy-pasting URLs into a separate dashboard.
How do I install vibekit-mcp?
Run `npm install -g vibekit-mcp` then add it to your Claude Desktop / Cursor MCP config. For Claude Desktop, edit ~/Library/Application Support/Claude/claude_desktop_config.json and add vibekit-mcp under mcpServers. The server uses your VibeKit API key for auth, get one from app.vibekit.bot/settings.
Which clients support this?
Any client that implements the Model Context Protocol spec, currently Claude Desktop, Cursor, Zed, Continue, and a growing list. The server uses stdio transport so it works with the standard MCP config block any of them accept.
Is the MCP server free?
Yes, the npm package is free and open-source. VibeKit's free tier (1 hosted app) is enough to start. The MCP server is just a thin RPC layer; your usage is bounded by your VibeKit plan and your client's own model costs.