fastllm Model Context Protocol (MCP) Server

GitHub

Fastllm是后端无依赖的高性能大模型推理库。同时支持张量并行推理稠密模型和混合模式推理MOE模型，任意10G以上显卡即可推理满血DeepSeek。双路9004/9005服务器+单显卡部署DeepSeek满血满精度原版模型，单并发20tps；INT4量化模型单并发30tps，多并发可达60+。

5,112 stars

C++ 1 file · ~126,242 tokens

Devices (fastllm) MCP Server

Model Context Protocol (devices) server

View all fastllm servers

Ready-to-Use Client Configurations

{
  "mcpServers": {
    "devices": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-devices"
      ]
    }
  }
}

{
  "mcpServers": {
    "devices": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-devices"
      ]
    }
  }
}

{
  "mcpServers": {
    "devices": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-devices"
      ]
    }
  }
}

{
  "mcpServers": {
    "devices": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-devices"
      ],
      "disabled": false,
      "autoApprove": []
    }
  }
}

{
  "mcpServers": {
    "devices": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-devices"
      ]
    }
  }
}

docker run -i --rm mcp/devices