Salad MCP Server

A Model Context Protocol (MCP) server for SaladCloud - the GPU provider with super nice features!

This MCP server provides seamless integration with SaladCloud's API, enabling Claude and other MCP clients to manage GPU workloads, container groups, inference endpoints, and more.

Features

• Container Groups Management: Create, list, update, start, stop, and delete container groups
• Inference Endpoints: List and manage inference endpoints
• Queue Operations: Manage message queues for async workload processing
• Logs & Monitoring: Access system logs and workload errors
• Quotas: Check organization resource quotas
• LLM-as-Judge Auto-Eval: Comprehensive integration testing using Claude API

Installation

Quick Start (Recommended)

Use npx to run without installation:

npx salad-mcp-server

Via Smithery

To install with Smithery:

npx -y @smithery/cli install salad-mcp-server --client claude

From npm

npm install -g salad-mcp-server
salad-mcp-server

From Source

npm install
npm run build

Configuration

Set your SaladCloud API key as an environment variable:

export SALAD_API_KEY=your_api_key_here
export SALAD_ORG_NAME=your_organization_name

For auto-eval testing, also set:

export CLAUDE_API_KEY=your_claude_api_key_here

Usage

As MCP Server

Add to your MCP client configuration (e.g., Claude Desktop):

Using npx (recommended):

{
  "mcpServers": {
    "salad": {
      "command": "npx",
      "args": ["salad-mcp-server"],
      "env": {
        "SALAD_API_KEY": "your_api_key_here",
        "SALAD_ORG_NAME": "your_organization_name"
      }
    }
  }
}

Using local installation:

{
  "mcpServers": {
    "salad": {
      "command": "node",
      "args": ["/path/to/salad_mcp/dist/index.js"],
      "env": {
        "SALAD_API_KEY": "your_api_key_here",
        "SALAD_ORG_NAME": "your_organization_name"
      }
    }
  }
}

Run Auto-Eval CLI

Test the full integration with LLM-as-judge evaluation:

npm run eval

Run Integration Tests

npm test

Available Tools

• list_container_groups: List all container groups in an organization/project
• get_container_group: Get details of a specific container group
• create_container_group: Create a new container group
• update_container_group: Update an existing container group
• start_container_group: Start a container group
• stop_container_group: Stop a container group
• delete_container_group: Delete a container group
• list_inference_endpoints: List available inference endpoints
• get_quotas: Get organization resource quotas
• list_queues: List message queues
• create_queue: Create a new message queue
• get_queue: Get queue details
• delete_queue: Delete a queue

API Reference

Based on SaladCloud API Documentation

Sources

• SaladCloud API Documentation
• Container Groups API
• SaladCloud Python SDK

License

MIT