Teach any AI agent (Claude, Cursor, Codex, …) to drive Xcelsior GPU compute through the official TypeScript SDK — no MCP client required.
npx skills add xcelsior-gpu/skillThen set your access token (from https://xcelsior.ca/dashboard/mcp):
export XCELSIOR_ACCESS_TOKEN="xcel_ai_..."The agent installs @xcelsior-gpu/sdk
and can then list GPUs, check spot pricing, estimate cost, launch and monitor
instances, and check your wallet balance — all in natural language.
The SDK is typed, exposes real error classes (XcelsiorApiError,
XcelsiorApiTimeoutError), and is generated from the same OpenAPI spec that backs the
public API — so an agent using it cannot drift from the wire format the way hand-written
requests do.
Xcelsior also ships a hosted MCP server at https://xcelsior.ca/mcp. See the
dashboard's Connect page for a one-paste setup with your token embedded.
Full API reference: https://docs.xcelsior.ca