Ways to use Superjolt
Skill teaches the agent when to use Superjolt; MCP is the tools it calls; the API is for CI/scripts; the dashboard is read-mostly. All point at the same control plane.
Superjolt is one control plane with a few front doors. They all act on the same VMs, domains, and account — pick whichever fits the moment.
Skill — the discovery layer
A small instruction file that teaches your agent when to reach for Superjolt, so it provisions real compute on its own. It doesn’t carry the tools — it points at the MCP server. Start here.
MCP — the tools
The Superjolt MCP server (https://mcp.superjolt.com) is where the tools live — create_vm, add_domain, open_port, and the rest. Any MCP-capable agent (Claude Code, Cursor, Claude Desktop, …) connects to it and calls them directly. This is the primary way an agent operates Superjolt.
→ Getting started · MCP tools reference
API — for CI and scripts
Everything the agent does is a REST call under the hood. For automation without an agent — a deploy step, a cron job — hit https://api.superjolt.com/v1/* directly with an API token (Authorization: Bearer sj_live_…). Mint a token from the dashboard.
Dashboard — read-mostly
The dashboard is where you watch what’s running, see costs, and mint API tokens. It’s deliberately read-mostly — provisioning and changes happen through the agent (MCP), not by clicking around.