skiff-cloud

Sovereign Bare-Metal Cloud CLI, AI Agent SDK & Model Context Protocol (MCP) Server

skiff-cloud lets developers and AI agents (Claude, Cursor, LangChain, Vercel AI SDK) inspect, deploy, and interact directly with sovereign bare-metal infrastructure on node ind-tbn-1 (Agnirockz).

1. Quickstart (CLI)

Run directly with npx or install globally:

# Run without installing
npx skiff-cloud --help

# Or install globally
npm install -g skiff-cloud

Agent Auto-Setup (Cursor / Claude)

Run a single command in any project directory:

npx skiff-cloud init

This automatically configures .cursor/mcp.json with 6 sovereign bare-metal tools!

Authentication & Identity

# Authenticate with Sovereign Node
skiff login
# (Session cached securely to ~/.skiff/config.json)

# Check active tenant, user role, and cgroups v2 hardware quotas
skiff whoami

# Manage endpoints and mesh connections
skiff config get
skiff config set-endpoint http://100.95.206.7:8090

# Mint and list AWS SigV4 keys for S3 storage
skiff keys list
skiff keys create --name "agent-key" --role developer

Bare-Metal Primitives

# Inspect real-time 6-core CPU, RAM, NVMe & GPU telemetry
skiff status

# List and deploy services (Git repos or images)
skiff ps
skiff deploy --repo vercel/next-learn --name my-app
skiff deploy --image nginx:alpine --port 80 --name web

# Run SQL on tenant PostgreSQL + pgvector
skiff db query "SELECT * FROM documents LIMIT 5;"

# Stream local completions from the NVIDIA GPU
skiff chat "Explain LSM-Tree memtable mechanics"

2. Connect Your AI Agent via MCP (Model Context Protocol)

skiff-cloud includes a built-in MCP server that enables AI agents in Cursor, Claude Desktop, or custom agent loops to control your sovereign hardware tools.

1-Click Setup:

npx skiff-cloud init

Manual Claude Desktop Configuration

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "skiff": {
      "command": "npx",
      "args": ["-y", "skiff-cloud", "mcp"]
    }
  }
}

Tools Exposed to AI Agents:

• skiff_telemetry — Inspect CPU load, DDR4 memory, NVMe disk, and GPU thermals.
• skiff_list_sandboxes — List all active Docker container microservices and statuses.
• skiff_deploy_sandbox — Provision and start a new container sandbox with subdomain ingress.
• skiff_sql_query — Execute SQL queries against PostgreSQL + pgvector.
• skiff_gpu_chat — Query local NVIDIA GPU silicon with zero third-party data leakage.
• skiff_list_keys — List active AWS SigV4 storage credentials.

3. Programmatic TypeScript Agent SDK

Integrate directly into Node.js / TypeScript agents:

import { SkiffClient } from "skiff-cloud";

const skiff = new SkiffClient({
  endpoint: process.env.SKIFF_API_URL, // defaults to https://api.skiffcloud.app
  token: process.env.SKIFF_TOKEN, // Or loads automatically from ~/.skiff/config.json
});

// Deploy a service from a Git repo or image
const service = await skiff.deployService({
  name: "embedding-service",
  repo: "owner/repo",
  memoryMB: 1024,
});
console.log(`Deployed to: ${service.url}`);

// Query PostgreSQL with pgvector
const results = await skiff.queryDatabase("SELECT id, title FROM articles ORDER BY embedding <-> '[0.1, 0.2]' LIMIT 5;");
console.log(results.rows);

// Run local inference on the bare-metal GPU
const answer = await skiff.chatCompletion("Summarize this document");
console.log(answer);

4. Ingress & Network Modes

• Public Cloudflare Gateway: Out-of-the-box zero-setup public ingress (https://api.skiffcloud.app).
• Tailscale Encrypted Direct Mesh: Sub-millisecond latency for operators on the mesh via http://100.95.206.7:8090 (set explicitly with skiff config set-endpoint).
• Custom Endpoints: Switch endpoints anytime using npx skiff-cloud config set-endpoint <url>.