Connecting a coding agent to a cloud usually goes like this: create an API key, find the right config file, paste in a JSON block, and restart the agent.
Our setup guide is one sentence. Paste it into your agent:
“Read agents.runpod.io and set up Runpod for me.”
Your agent installs our skills—tested procedures for jobs like deploying a Pod—and connects our hosted MCP server, which carries current docs. Then you sign in with your Runpod account in your browser, or use an API key if your client doesn’t support browser sign-in.
Every MCP tool is labeled read-only or destructive, so you decide which actions your agent takes without asking you first.
One-prompt setup is part of the November release of Runpod for Agents, and more is coming in December. It works in 17 agent clients, including Claude Code, Codex, Cursor and ChatGPT.
Our new Model Hub lists 43 models you can call through an API today. Each model’s page also shows the GPU it runs on, the code to deploy it and what it costs per hour to run it on your own endpoint.
Weld started as a weekend experiment and now builds and pushes container images from inside a Pod, no Docker daemon required. It's still alpha, so bugs go to GitHub.
A public template shows every field to everyone, environment variables included. A private template doesn't make its image private, either. Keep credentials in Runpod secrets, where they stay yours.
Package a PyTorch image with your code, save it as a private template and launch a Pod from it. The finish line is your GPU returning [2.0, 4.0, 6.0], which feels better than it should.
Virtual: Build long-running agents with Kimi — November 12
How do you keep a long-running Kimi agent from eating your token budget? Moonshot AI's DevRel lead covers which Kimi model fits which job and how to cut usage.