MCP Integration

Give your AI coding agent the ability to spin up burners. When the agent needs to show you a live preview, share a URL with a third party, or verify its work in a real environment — it just calls spin_up_burner and gets a public URL back in ~30 seconds.

Setup

Two steps:

  1. Authenticate (one-time)
    npx burner-cli login

    Signs in via GitHub OAuth and stores your API key at ~/.config/burner/config.json.

  2. Add to your agent's MCP config

Claude Code

Add to your project's .mcp.json or global settings:

{
  "mcpServers": {
    "burner": {
      "command": "npx",
      "args": ["burner-cli", "mcp"]
    }
  }
}

Cursor

Add to .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "burner": {
      "command": "npx",
      "args": ["burner-cli", "mcp"]
    }
  }
}

Other MCP clients

Any MCP client that supports stdio transport can use the burner server. Point it at npx burner-cli mcp as the command.

Available Tools

Once configured, your agent will see these tools:

spin_up_burner

The main tool. Deploys a local project directory to a live public URL. The agent provides a directory path and a start command, and gets back a URL like https://cool-fire-abc.preview.burner.dev.

ParameterRequiredDescription
directoryYesPath to the project directory
commandYesStart command (e.g. npm start)
portNoServer port (default: 3000)
envNoEnvironment variables as key-value pairs
ttl_minutesNoTime-to-live in minutes (default: 45)

list_burners

Lists all your burners. Optionally filter by status (running, failed, destroyed, etc).

get_burner

Gets the current status and details of a specific burner by its preview ID.

destroy_burner

Tears down a burner and its cloud resources. Burners also auto-destroy when their TTL expires, so this is optional.

When Will an Agent Use This?

The tool descriptions are designed so your agent understands when to reach for a burner on its own:

  • You're away from your machine — the agent is building a feature and wants to show you progress. It spins up a burner and sends you the link.
  • No localhost available — your agent is running in a cloud environment without port forwarding. A burner gives it a real URL to work with.
  • Sharing with others — you want to show a colleague or client the current state of the app before committing anything.
  • Quick verification — the agent wants to confirm its changes actually work in a deployed environment, not just locally.

How It Works Under the Hood

When the agent calls spin_up_burner, the MCP server:

  1. Archives the project directory (respects .gitignore)
  2. Uploads the archive to cloud storage
  3. Boots an isolated VM with your code
  4. Runs dependency install + your start command
  5. Returns a live public URL

The entire process takes ~30 seconds. The agent gets back the URL along with the preview ID (for later get_burner or destroy_burner calls).