fiber

本地服务

fiber.ac云与基础设施

代码仓库访问官网

关于 fiber

面向 AI 数据中心的光学运维 MCP：gNMI、ML RUL、数字孪生、NCCL、SPARQL 语义查询

配置

本地服务的配置在各客户端中通用，复制后粘贴到客户端的 MCP 配置文件即可。

mcp.json · 通用配置

{
  "mcpServers": {
    "liveplex-cpu-fiber": {
      "command": "uvx",
      "args": [
        "fiber-mcp"
      ],
      "env": {
        "FIBER_API_URL": "<FIBER_API_URL>",
        "FIBER_API_KEY": "<FIBER_API_KEY>",
        "FIBER_ENABLE_WRITE": "<FIBER_ENABLE_WRITE>",
        "FIBER_MCP_TOOLS_ALLOWLIST": "<FIBER_MCP_TOOLS_ALLOWLIST>",
        "FIBER_HTTP_TIMEOUT": "<FIBER_HTTP_TIMEOUT>"
      }
    }
  }
}

启动命令uvx fiber-mcp

需要配置的环境变量

FIBER_API_URLBase URL of the FIBER deployment (e.g., https://api.fiber.ac or http://localhost:8000 for self-hosted).

FIBER_API_KEY密钥Bearer token for the FIBER API. Role determines which tools work: viewer (24 read tools), operator (+write), admin (all).

FIBER_ENABLE_WRITESet to 1 to expose write tools: demo trigger, NCCL ingest, RMA create. Defaults to read-only for safety.

FIBER_MCP_TOOLS_ALLOWLIST(v1.0.2) Optional CSV of tool names. When set, only these tools are exposed — letting operators scope autonomous agents to a minimum surface (e.g. 'fiber_health,fiber_cluster_overview,fiber_list_alerts'). Empty = all tools exposed (default).

FIBER_HTTP_TIMEOUTPer-request HTTP timeout in seconds. Increase if running Digital Twin Monte Carlo with iterations>1000.

本地服务需要在你的电脑上安装运行环境（如 Node.js 18+）。各客户端的配置文件位置不同，请参考客户端文档。

安装包

pypifiber-mcp1.1.0stdio