Runpod Plugins: GPU cloud for agents
Runpod Official Plugins
The Runpod plugin with 16 skills and an MCP server: an agent starts pods, deploys endpoints and audits GPU spend
- 48 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The agent creates and deletes pods, endpoints and volumes, and the spend comes from the Runpod balance
- One API key opens both MCP and the command-line tools, and a saved key stays in `~/.runpod/config.toml`
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/runpod-official-pluginsDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent creates and deletes pods, endpoints and volumes, and the spend comes from the Runpod balance
- One API key opens both MCP and the command-line tools, and a saved key stays in `~/.runpod/config.toml`
This is third-party code. Review the repository files before installing.
What it does
The plugin bundles 16 skills and a connection to the hosted Runpod MCP server under one manifest. The entry point is the `runpod` skill: it picks how to run the task, through MCP tools, the `runpodctl` command line or the `flash` Python library. Through them the agent starts GPU and CPU pods, deploys serverless endpoints and edits templates and network volumes. Step-by-step MCP journeys cover resource discovery, create and delete, serverless and pod deployment, pod diagnosis, endpoint operations and cost audit. Other skills explain how Runpod works, the ready-made pod templates, building your own template and moving code from the GraphQL API and REST v1 to REST v2. The golden-paths folder holds about two dozen examples verified on a live account. They have no SKILL.md, so the agent opens them when asked something like "look at the examples".
Who it is for. Developers and ML engineers who keep models and workers on Runpod and want to manage pods and endpoints from a conversation with an agent.
Good fit when
- An experiment needs an A100 pod and you want to start it with one sentence to the agent
- An endpoint keeps returning IN_QUEUE and never processes requests
- A Python handler has to go to serverless through flash
- The GPU bill grew and you want a cost audit of pods and volumes
- Code was written for the Runpod GraphQL API and has to move to REST v2
Not a fit when
- The workload runs in another cloud: the plugin knows only Runpod
- There is no Runpod account or API key, and MCP and `runpodctl` need them
- You only want to understand how Runpod works: the `runpod-usage` skill or the documentation is enough without account access
- Pods and volumes should not be deleted without reviewing a plan: ask for the plan before changes
Example request
List my Runpod endpoints and find the most expensive pod that sits idle.Limitations
`runpodctl` and `flash` need `RUNPOD_API_KEY`, and you create the key in the Runpod console. The MCP server accepts an OAuth sign-in or the same key in a Bearer header. Codex may not wire the MCP server itself, in which case add it with `codex mcp add runpod --url https://mcp.runpod.io/`. The skills call themselves a snapshot: check capabilities and syntax against the tools. The `docker`, `gh`, `hf` and `aws` utilities install separately per task.
How to disable. In Claude Code run `/plugin uninstall runpod@runpod`, then `/plugin marketplace remove runpod`: the MCP registration goes with the marketplace. Skills from skills.sh are removed with `npx skills remove runpod`.
FAQ
How do I install the Runpod plugin in Claude Code?
Run `/plugin marketplace add runpod/runpod-plugins-official`, then `/plugin install runpod@runpod` and `/reload-plugins`. The hosted MCP server connects together with the plugin, and you sign in through `/mcp`, the runpod entry.
Do I need an API key?
`runpodctl` and `flash` need `RUNPOD_API_KEY` from the Runpod console. The MCP server does not require a key: an OAuth sign-in with nothing on disk is enough. The key also works for MCP when passed in a Bearer header.
How do I install the plugin in Cursor or Codex?
In Cursor, Copilot, Windsurf, Cline, opencode and Gemini the skills install with `npx skills add runpod/runpod-plugins-official` and MCP is added with `npx @runpod/mcp-server@latest add`. In Codex use `codex plugin marketplace add https://github.com/runpod/runpod-plugins-official.git` and `codex plugin add runpod@runpod`.
Should I pick MCP or runpodctl?
MCP gives structured tool calls when the server is connected in the session. `runpodctl` fits the terminal, CI and scripts, and it also transfers files and manages SSH keys. The `runpod` router picks the path for the task.
How do I update the Runpod plugin?
In Claude Code run `/plugin marketplace update runpod`, then `/reload-plugins`. In Codex use `codex plugin marketplace upgrade runpod`, and for skills.sh skills re-run `npx skills add runpod/runpod-plugins-official`.
Discussion
No questions or reviews yet.
Related
PluginsOfficial Runpod MCP server: an agent manages Pods, Serverless endpoints, templates and network volumes from chat.
AWS's official set of MCP servers, skills and plugins so an agent can build, deploy and operate on AWS
An AWS marketplace of plugins that give the agent skills to architect, deploy and operate AWS services
Azure Skills
Azure Skills Plugin
Microsoft's official plugin: Azure skills plus the Azure MCP Server and Foundry MCP wired in for real cloud actions
The vp run command starts Claude Code, Codex, Gemini and other CLI agents in Docker or Podman containers
Google's MCP server for deploying apps to Cloud Run from an agent, with service listing and log reading
README in short
The repository ships a single `runpod` plugin. One manifest serves Claude Code, Codex, Gemini and opencode, and through skills.sh the same skills reach Cursor, Copilot, Windsurf, Cline and other agents. Each release also publishes the npm package `@runpod/plugin-knowledge`, the plugin content as one JSON file for programs.