Lingie Logo灵姬 LingieLingieAI.com

MCP Integration

Let AI agents directly call Lingie
Generate images, videos, and audio

Lingie's official MCP server (Lingie.Mcp) wraps the client's generation capabilities as standard MCP tools. Any MCP-compatible AI agent, such as Claude Desktop, Cursor, and ZCode, can connect in one click: write a prompt, and leave the rest to the agent.

View GitHub repository ↗Free Client Download

Why Lingie MCP

Zero-dependency single file

A pure Node.js (>=18) single-file implementation. No npm install needed; clone and run.

Full coverage: local + cloud

Free local ComfyUI workflows, or call cloud GPUs and third-party models such as DALL·E, Midjourney, and Kling through the gateway.

Long tasks never lose results

If not completed in about 20 seconds, it automatically falls back and returns a task ID; the agent polls to retrieve the result, unaffected by the host's 30-second call timeout.

Zero-modification integration

Directly reuse Lingie's built-in local API; gateway credentials are issued with one click in Lingie settings and take effect automatically in about 10 seconds.

Quick start

Three steps to integrate: clone the repository, copy the config, and restart the agent to see the lingie tools.

1. Prerequisites

  1. 1灵姬客户端正在运行（官网免费下载）
  2. 2灵姬 设置 → 本地 API → 开启「外部工作流 API」（端口默认 17856）
  3. 3无需手动启动 ComfyUI 引擎：提交任务时灵姬会自动拉起（首次约 30-120 秒）

2. Clone the repository

git clone https://github.com/MechaBabyAi/Lingie.Mcp.git C:\MCP\Lingie.Mcp
GitHub ↗

3. Configure the agent

Add the following to the MCP configuration file of ZCode / Claude Desktop / Cursor (replace the path with your actual clone directory):

{
  "mcpServers": {
    "lingie": {
      "command": "node",
      "args": ["C:\\MCP\\Lingie.Mcp\\server.mjs"]
    }
  }
}

Tool list

By default, the Workflow API tool is exposed (local ComfyUI engine, free):

lingie_health检查灵姬本地 API 与本地 ComfyUI 引擎是否在线（未运行时提交会自动拉起）
lingie_list_workflows列出可用工作流（图片/视频/音频生成等）
lingie_workflow_capabilities查看某工作流的参数模式（类型/必填/默认值/可选值/提交方式）
lingie_upload_file上传本地文件到 ComfyUI 输入目录（供图生图/参考图/首帧等参数使用）
lingie_run_workflow执行工作流：完成直接返回输出文件本地路径，超过约 20 秒自动降级返回 run_id
lingie_run_result轮询 run_id 并取回结果，完成时返回输出文件本地路径
lingie_run_status查询任务状态与进度（不下载输出）
lingie_cancel_run取消任务（会中断本地引擎当前正在执行的任务）

When gateway credentials are available (Lingie Settings → Local API → enable Agent (MCP) Gateway Access, or set environment variables), additional gateway tools are exposed, covering local/cloud ComfyUI and third-party models (billed by Lingie credits):

lingie_list_models列出网关可用模型（本地/云端 ComfyUI 与第三方模型）
lingie_generate_image一键调用模型生成图片（超时自动降级返回 task_id）
lingie_generate_video一键调用模型生成视频（超时自动降级返回 task_id）
lingie_task_status查询网关任务状态与输出
lingie_cancel_task取消网关任务
lingie_estimate_cost预估一次模型调用的灵力值消耗（不实际生成）
lingie_spirit_balance查询灵力值余额（可用 = 余额 − 冻结）

Port, token, output directory, logs, etc. can all be overridden via environment variables; see the repository README for details. After enabling gateway access on the Lingie side, credentials are automatically written to local mcp-credentials.json, no manual configuration needed.

Typical agent invocation flow

Take text-to-image as an example: a complete generation takes only 6 steps:

1lingie_health
确认灵姬在线（引擎未运行也可提交，会自拉起）
2lingie_list_workflows
找到想用的工作流（如「文生图」）
3lingie_workflow_capabilities
了解参数怎么填
4lingie_upload_file（可选）
图生图先上传参考图
5lingie_run_workflow
提交生成：约 20 秒内完成直接拿到输出文件本地路径，否则拿到 run_id 走下一步
6lingie_run_result
轮询 run_id，完成时返回输出文件本地路径，Agent 直接处理

Ready to let the agent do it for youCreateReally?

Clone Lingie.Mcp and get integrated in minutes; generation runs entirely through the Lingie client, with transparent and controllable Lingie points billing.

Go to GitHubDownload Lingie first

LingieAI.com