Kilawatt MCP server

kilawatt-mcp-server

Model Context Protocol (MCP) server for Kilawatt Cloud GPU provisioning. Enables Cursor, Claude Desktop, and other AI agents to programmatically request and manage real GPU compute nodes.

This server calls the real Kilawatt gateway — every response is an actual HTTP result from https://www.kilawattcloud.dev/api/public/v1. There are no mock or fabricated responses.

Installation

Install via npm:

npm install kilawatt-mcp-server

Configuration

Set your Kilawatt API key as an environment variable before running the server:

export KILAWATT_API_KEY=kw_live_YOUR_KEY_HERE

The API key must start with kwlive. API keys are issued through your Kilawatt Cloud account. Contact hello@kilawattcloud.dev for assistance if you don't have access to generate keys yet.

Optionally, override the gateway URL:

export KILAWATT_BASE_URL=https://www.kilawattcloud.dev/api/public/v1

Running the Server

Start the MCP server:

npm start

Or directly:

node index.js

The server listens on stdio for MCP requests from your agent.

Available Tools

deploygpunode

Provision real GPU compute on Kilawatt Cloud. This immediately charges the caller's Kilawatt wallet and starts an actual machine. Set dry_run: true to validate pricing, routing, and account balance without provisioning.

Parameters:

Example: Deploy 4 H100 GPUs for 1 hour at lowest cost

{
  "name": "deploy_gpu_node",
  "arguments": {
    "gpu_type": "nvidia-h100",
    "card_count": 4,
    "duration_seconds": 3600,
    "routing_policy": "lowest_cost",
    "dry_run": false
  }
}

Example: Dry-run to check pricing

{
  "name": "deploy_gpu_node",
  "arguments": {
    "gpu_type": "nvidia-b200",
    "card_count": 2,
    "duration_seconds": 1800,
    "dry_run": true
  }
}

Response (Success):

GPU node running. Job ID: job_abc123xyz
Node: provider-name
Hardware: 4× nvidia-h100
Workload: default   Routing policy: lowest_cost
Billed: $12.3456
Status: running

Response (Dry Run):

DRY RUN — nothing was provisioned and nothing was charged.
Authorized: yes
Estimated cost: $6.1728 for 2× nvidia-b200
Routing policy: lowest_cost
Would route through (in order): provider-a → provider-b → provider-c

Error Handling

All errors are descriptive and indicate whether anything was provisioned or charged. Common scenarios:

Shortened. The full README is on GitHub.