Runpod MCP server

This Model Context Protocol (MCP) server lets you manage Runpod infrastructure through any MCP-compatible client. It provides tools for working with Pods, Serverless endpoints, templates, network volumes, and container registry authentications.

Quick start

Requirements

• Node.js 18 or higher.
• A Runpod account and API key (get your API key).

Running with npx

You can run the server directly without installation:

RUNPOD_API_KEY=YOUR_API_KEY npx @runpod/mcp-server@latest

Installing via Smithery

To install for Claude Desktop automatically via Smithery:

npx -y @smithery/cli install @runpod/runpod-mcp-ts --client claude

Setting up with your client

Most MCP clients use a JSON configuration file with the same general structure. The examples below show the npx approach (recommended for most users) and the local build approach (for development). Replace YOUR_API_KEY with your actual Runpod API key.

Claude Code

Add the MCP server globally so it's available across all your projects:

claude mcp add runpod -s user \
  -e RUNPOD_API_KEY=YOUR_API_KEY \
  -- npx -y @runpod/mcp-server@latest

Or add it to a specific project (creates a .mcp.json file you can commit):

claude mcp add runpod -s project \
  -e RUNPOD_API_KEY=YOUR_API_KEY \
  -- npx -y @runpod/mcp-server@latest

Verify the server is connected with claude mcp list. If you're in an active session, type /mcp to reconnect without restarting.

Claude Desktop

Edit the config file at ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "runpod": {
      "command": "npx",
      "args": ["-y", "@runpod/mcp-server@latest"],
      "env": {
        "RUNPOD_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

Restart Claude Desktop after saving.

Cursor

Add the following to .cursor/mcp.json in your project directory, or ~/.cursor/mcp.json for global access:

{
  "mcpServers": {
    "runpod": {
      "command": "npx",
      "args": ["-y", "@runpod/mcp-server@latest"],
      "env": {
        "RUNPOD_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

Windsurf

Open Windsurf settings (Cmd+Shift+P → "Open Windsurf Settings"), navigate to the Cascade section, and enable MCP. Then edit ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "runpod": {
      "command": "npx",
      "args": ["-y", "@runpod/mcp-server@latest"],
      "env": {
        "RUNPOD_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

VS Code (GitHub Copilot)

MCP works in VS Code's agent mode (requires VS Code 1.101+). Add the following to .vscode/mcp.json in your workspace:

{
  "mcpServers": {
    "runpod": {
      "command": "npx",
      "args": ["-y", "@runpod/mcp-server@latest"],
      "env": {
        "RUNPOD_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

Click the "Start" button next to the server entry to connect.

Cline

Open Cline in VS Code, click the hamburger menu (☰), and go to MCP Servers. You can add servers through the marketplace or manually configure in Cline's settings using the same JSON structure shown above.

JetBrains IDEs

Create a mcp.json file at ~/.junie/mcp.json (global) or .junie/mcp/ in your project:

{
  "mcpServers": {
    "runpod": {
      "command": "npx",
      "args": ["-y", "@runpod/mcp-server@latest"],
      "env": {
        "RUNPOD_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

Other clients

This server uses stdio transport and works with any MCP-compatible client. The configuration pattern is the same across all clients — point the command to npx with @runpod/mcp-server@latest as the argument, and set RUNPOD_API_KEY in the environment. For a full list of MCP clients, see the official MCP clients page.

Using a local build

If you want to run from a local clone of the repo (for development or to test unreleased changes):

git clone https://github.com/runpod/runpod-mcp.git
cd runpod-mcp
pnpm install
pnpm build

Then replace the command and args in any of the configurations above with:

{
  "command": "node",
  "args": ["/absolute/path/to/runpod-mcp/dist/index.mjs"]
}

After making changes to the source, re-run pnpm build and restart or reconnect the MCP server for changes to take effect.

Usage examples

List all Pods

Can you list all my Runpod Pods?

Create a new Pod

Create a new Runpod Pod with the following specifications:
- Name: test-pod
- Image: runpod/pytorch:2.1.0-py3.10-cuda11.8.0-devel-ubuntu22.04
- GPU Type: NVIDIA GeForce RTX 4090
- GPU Count: 1

Create a Serverless endpoint

Create a Runpod Serverless endpoint with the following configuration:
- Name: my-endpoint
- Template ID: 30zmvf89kd
- Minimum workers: 0
- Maximum workers: 3

Contributing

To get started with local development, clone the repo and build:

git clone https://github.com/runpod/runpod-mcp.git
cd runpod-mcp
pnpm install
pnpm build

After making changes, rebuild with pnpm build. In Claude Code, type /mcp to reconnect to the updated server without restarting your session. You can also use pnpm build:watch for auto-rebuilding during development.

All tools live in src/index.ts. The server uses two backends: the REST API (runpodRequest()) for authenticated CRUD operations, and the GraphQL API (graphqlRequest()) for public read-only queries like GPU types and data centers. Follow existing patterns when adding new tools — kebab-case names, Zod schemas with .describe(), and JSON stringified responses.

This project uses changesets for versioning and npm publishing. Every PR with user-facing changes needs a changeset file at .changeset/DESCRIPTIVE_NAME.md:

---
"@runpod/mcp-server": minor
---

Description of what changed and why.

Use patch for bug fixes, minor for new tools or features, and major for breaking changes. The .changeset/ directory is gitignored, so use git add -f to stage changeset files.

See CLAUDE.md for the full development guide including architecture details, tool conventions, and known issues.

Security considerations

This server requires your Runpod API key, which grants full access to your Runpod account. Never share your API key. Be cautious about what operations you perform, and consider setting up a separate API key with limited permissions. Do not use this in a production environment without proper security measures.

License

Apache-2.0