
varity-mcp
io.github.varity-labs/varity-mcp
Deploy apps and AI agents to production from your AI editor, with the cost shown up front.
Documentation
@varity-labs/mcp
The published Varity MCP package for Cursor, Claude Code, VS Code, Windsurf, and other MCP clients.
This repository contains only @varity-labs/mcp, the npm MCP server. It is a thin client of Varity's public API; it is not the Python CLI, the portal, the dormant SDK, or an app-store package.
The Varity MCP Server lets a supported local AI coding client build, deploy, and manage supported apps in production for you.
Ask your AI to put your app online: you can see the cost before anything deploys, a static site is usually live in seconds, and your keys stay in your project's .env file, never in the chat.
Quick start: pick your editor below and run one command.
Install
One-click install
GitHub does not open editor links, so use the install buttons on the Varity MCP docs page, or open the link for your editor from your browser's address bar:
Cursor: cursor://anysphere.cursor-deeplink/mcp/install?name=varity&config=eyJjb21tYW5kIjoibnB4IiwiYXJncyI6WyIteSIsIkB2YXJpdHktbGFicy9tY3BAbGF0ZXN0Il19
VS Code: vscode:mcp/install?%7B%22name%22%3A%22varity%22%2C%22command%22%3A%22npx%22%2C%22args%22%3A%5B%22-y%22%2C%22%40varity-labs%2Fmcp%40latest%22%5D%7D
VS Code Insiders: vscode-insiders:mcp/install?%7B%22name%22%3A%22varity%22%2C%22command%22%3A%22npx%22%2C%22args%22%3A%5B%22-y%22%2C%22%40varity-labs%2Fmcp%40latest%22%5D%7D
Each link installs npx -y @varity-labs/mcp@latest and carries no key: you connect your Varity account afterwards with varity_login.
Cursor
Add to .cursor/mcp.json:
{
"mcpServers": {
"varity": {
"command": "npx",
"args": ["-y", "@varity-labs/mcp@latest"]
}
}
}
Claude Code
claude mcp add varity -- npx -y @varity-labs/mcp@latest
VS Code with Copilot
- Command Palette → MCP: Add Server
- Select Command (stdio)
- Command:
npx -y @varity-labs/mcp@latest - Name:
Varity
Windsurf
Add to mcp_config.json:
{
"mcpServers": {
"varity": {
"command": "npx",
"args": ["-y", "@varity-labs/mcp@latest"]
}
}
}
Any MCP-compatible client (stdio)
{
"mcpServers": {
"varity": {
"command": "npx",
"args": ["-y", "@varity-labs/mcp@latest"]
}
}
}
What you can do from your AI editor
The MCP exposes Varity's deploy surface so your AI tool can take action — not just generate code. Try prompts like:
Deploy your own code
- "Deploy this project to production"
- "Ship this app live"
- "What would it cost to deploy this on Varity?"
Deploy a supported public template (one command, no code required)
- "What templates can Varity deploy?"
- "Which templates need no secrets?"
- "Deploy this public, secret-free template for me"
Manage live deployments
- "Show my deployments"
- "What's the URL of my last deploy?"
- "Stop my-app and stop billing it"
- "Why did my deployment fail?"
- "Show the runtime logs for my deployment"
Preview a Vercel migration
- "Preview what would change when migrating github.com/me/my-app"
- "Show me the Vercel-specific artifacts to replace"
When your app is live
Ask your agent to finish with one plain line: where the app is, what it can cost, and how to stop paying. It reads like this:
Your app is live at
<live URL>. It costs up to<monthly maximum>/month (monthly maximum). Traffic alone does not change an app's monthly price. To stop billing, ask me to delete it.
The URL comes from varity_deploy_status and the monthly maximum from varity_cost_calculator or varity_template_info. The traffic sentence applies to apps only; virtual machines are priced by the hour.
Docs and pricing
- "Search Varity docs for environment variables"
- "What is the current estimate for this pricing profile?"
- "What billing projection does my deployed app report?"
Tools
| Tool | What it does |
|---|---|
varity_search_docs | Search the Varity documentation |
varity_cost_calculator | Project the live owner's monthly or hourly USD estimate without converting units; pre-deploy estimates need no sign-in |
varity_balance | Read your available balance, the amount set aside for your running apps (reserved_usd) and the amount owed; the message rounds to cents and the data keeps the exact decimal USD strings; unavailable reads never become zero |
varity_doctor | Check that your environment is ready to deploy |
varity_login | Connect Varity: when browser sign-in is switched on, opens a browser tab where you confirm (or shows a short code on a machine without a browser); until then it tells you to add a deploy key to your editor's MCP settings and restart the editor |
varity_logout | Sign this computer out: revokes the saved sign-in on Varity, then removes it |
varity_install_deps | Install project dependencies |
varity_build | Build the project |
varity_open_browser | Open a URL locally |
varity_dev_server | Start the local development server |
varity_create_repo | Create a GitHub repository and push the project |
varity_deploy | Deploy the current project to production |
varity_deploy_status | Check the status of a deployment: the current step and time so far while it deploys, how long it took once live, and the last lines of the build error when a build failed |
varity_deploy_logs | Read runtime logs |
varity_delete_deployment | Request deletion; poll varity_deploy_status until it is deleted and billing is stopped |
varity_set_env | Set an app's settings (such as API keys) from the project's own .env file: you name the settings, the values are read on your computer and never pass through chat. Your app restarts; restarting holds one more month's maximum, released when billing next settles, usually within 20 minutes |
varity_domain_add | Point your own domain at an app and get the exact DNS record to add at your registrar |
varity_domain_verify | Check whether that DNS record is in place and the domain is live |
varity_domain_remove | Stop serving an app at its custom domain (it keeps its Varity address) |
varity_redeploy | Restart a deployed dynamic or container app from its last successful release at its saved commit; a static site cannot be restarted. To ship new commits, push and run varity_deploy again with the same app name |
varity_machines_list | List your CPU and GPU virtual machines that are still running or billing, one line each (include_stopped adds stopped ones; limit, default 20, up to 100), optionally with the CPU or, with gpu, GPU VM profile catalog |
varity_machines_create | Quote and create a CPU or (gpu: true) GPU virtual machine with your SSH public key |
varity_machines_delete | Delete a CPU or GPU virtual machine and confirm its billing stopped |
varity_list_templates | List certified gateway-owned Varity templates |
varity_template_info | Show full details for one certified template |
varity_deploy_template | Deploy a public certified template; a template's required settings are read from your project's .env file, never typed into chat; private templates are refused |
varity_list_agents | Backward-compatible alias for varity_list_templates |
varity_agent_info | Backward-compatible alias for varity_template_info |
varity_deploy_agent | Backward-compatible alias for varity_deploy_template |
varity_migrate | Preview Vercel-to-Varity source transformations without mutating or deploying |
Templates
Varity templates come from the gateway-owned certified catalog. Ask your AI editor "what templates can I deploy?" and it will list the live catalog and inspect each template contract. The MCP deploy tool supports public templates; when a template needs settings (for example an API key), put them in a .env file in your project folder and the tool reads them there, so the values never pass through chat. Private templates are refused.
Each template reserves different hardware. Use varity_template_info to see the required environment variables, private/public access mode, resources, hardware profile, and certification state before deciding whether the MCP can deploy it.
End-to-end example
From empty folder to deployed app, all in natural language:
You: "Make me a simple landing page for my coffee shop and deploy it"
AI: Wrote the landing page and ran its local build.
AI: Deploy accepted. Tracking its durable run until the owner reports a terminal outcome.
AI: The status owner now reports the app live at https://varity.app/coffee-shop/
You: "Now deploy Agent Zero"
AI: Agent Zero is certified and does not require environment variables.
AI: Template deploy accepted. No live URL is claimed until varity_deploy_status proves it.
Pricing
Pricing profiles and values are owned by Varity's live public interface. Ask
your AI editor to use varity_cost_calculator for a current estimate; this
package intentionally embeds no price table or billing-policy copy.
Transport
The package supports stdio for local MCP clients such as Cursor, Claude Code, VS Code, and Windsurf.
npx -y @varity-labs/mcp@latest
The former hosted endpoint and network transport are retired. Browser clients that require a remote MCP URL are not supported by this package.
Prerequisites
- Node.js >= 22.11 (the current supported LTS baseline; EOL Node 18/20 are unsupported)
- A Varity deploy key for deploys and account reads. This is the way to
connect today: create one at
https://developer.store.varity.so/dashboard/settings?tab=api-keys, add it
as
VARITY_DEPLOY_KEYin your editor's MCP settings for Varity, then restart your editor or AI assistant. Keep the key out of the chat. If the Portal asks for a payment method before it creates the key, you can still put a site online today without a card using Chat and Build: https://developer.store.varity.so/dashboard/ai-gateway?tab=chat&mode=build Connecting through a browser sign-in (varity_loginopens a browser tab, you confirm, and the sign-in is saved on this computer in~/.varitykit/config.json, readable only by you) is being switched on. Until it is,varity_loginsays so and points you to the deploy key. The key is read in this order:VARITY_DEPLOY_KEY, then the sign-in file~/.varitykit/config.json, then the legacyVARITY_API_KEY.varity_doctorreports which source supplied it. - varitykit (
pipx install varitykit) only forvarity_migrate
Deploy supported apps from your AI coding tool.
@varity-labs/mcpnpm install @varity-labs/mcp