// DOCUMENTATION

Documentation

MatMul rents GPU machines from the neocloud market: pick a live offer, run a plain VM or your Docker image, and SSH in, billed from one prepaid balance. Use the console, or script it with the CLI, the Python SDK, the REST API or an MCP client.

Get started

The Quickstart goes from sign-in to a shell on a GPU and back to zero spend in eight steps. In short:

pip install matmul-cloud
matmul login                              # approve in the browser (CI: MATMUL_API_KEY)
matmul billing add-credit --usd 20        # Stripe checkout link
matmul ssh-key add
matmul offers --gpu H100
matmul instance create --name dev --gpu H100 --max-price 3 --wait
matmul ssh dev
matmul instance rm dev                    # billing stops

GPU cloud

Quickstart: Sign in, add credit and an SSH key, launch a GPU, SSH in and terminate it.
CLI reference: Every matmul command and flag: offers, instances, SSH keys, billing and jobs.
Python SDK: import matmul: the same workflow from Python, standard library only.
REST API: The REST API behind all of it, with auth, endpoints, fields and errors.
MCP server: Find and launch GPUs from Claude Code, Claude Desktop, Cursor or VS Code with the matmul-mcp server (preview).
Pricing & billing: Live prices, the price guard, prepaid credit, new-account limits, freezes and the grace period.
Managed jobs: Run a command on the cheapest matching GPU and release the machine when it ends (coming soon).
Share an app from your GPU: Give ComfyUI, Jupyter or a dashboard on a MatMul GPU instance a public HTTPS link (coming soon).

SkyPilot attach

If you already run SkyPilot, you can keep it where it lives (your laptop, a VM, your cloud) and reach its dashboard from the MatMul console. A small connector on your box dials out to a private Cloudflare Tunnel, so there's nothing to open inbound and no VPN; the console proxies the dashboard to you and your teammates behind single sign-on. It's under Connections in the console sidebar, separate from renting GPUs.

Connect SkyPilot: Attach an existing SkyPilot API server (or deploy one) to the console over an outbound-only tunnel.
Manage & share: Open your SkyPilot dashboard through the tunnel, watch environment status, and share it with teammates.
Disconnect & remove: Disconnect a SkyPilot machine and tear down its tunnel: the two steps, and what's left behind.
Troubleshooting: Fixes for a SkyPilot environment stuck on Pending, a 403 dashboard, or a missing connector.

What's live and what isn't

MatMul is early, so here is where each piece stands.

Live: GPU and CPU instances, prepaid credit, SSH keys, teams with roles, the CLI, Python SDK and REST API, and SkyPilot attach.
Preview (works, but expect rough edges): the matmul-mcp server for MCP clients such as Claude Code and Cursor.
Coming soon: managed jobs with matmul run (built, not switched on yet) and public app links (built, switched off until abuse controls ship).
Not built yet: inference endpoints, app links behind your MatMul sign-in, and self-serve organization invites.

Help

Stuck? See Support for how to reach us, and Status for known problems.

← All docs