sdnext_mcp

sdnext_mcp

Enables LLM hosts to control SD.Next (Stable Diffusion Next) WebUI via 125 MCP tools for text-to-image, image-to-image, upscaling, control, tagging, model management, and more.

Category
访问服务器

README

sdnext-mcp

Model Context Protocol (MCP) server for SD.Next (Stable Diffusion Next) WebUI.

It wraps SD.Next's REST API (/sdapi/v1/* + /sdapi/v2/civitai/*) as 125 MCP tools so an LLM host (Claude Desktop, DSH, etc.) can drive every SD.Next usage scenario: text-to-image, image-to-image, control-guided generation, upscaling, detailer, LoRA, scripts, captioning/tagging/VQA, model & option management, server introspection, file/gallery, and CivitAI.

This is a thin HTTP client. It does not embed Python and does not launch the WebUI. You must run SD.Next yourself with --api --listen; the MCP server just talks to it over HTTP.

The core reason this exists: SD.Next's request parameters differ from A1111 (modular guidance, 1024 default canvas, built-in YOLO detailer, schedulers as a separate axis, UNet/DiT, etc.), so A1111 MCP code does not work against it. Every tool here is modeled on SD.Next's actual pydantic request fields.

Requirements

  • Node.js ≥ 18 (uses global fetch, FormData, AbortSignal.timeout)
  • A running SD.Next instance started with --api --listen (add --api-auth user:pass to enable auth)

Install

cd sdnext_mcp
npm install          # installs @modelcontextprotocol/sdk + zod

Run

Default is stdio transport (spawned by your MCP host):

# in your MCP host config
{
  "mcpServers": {
    "sdnext": {
      "command": "node",
      "args": ["/abs/path/to/sdnext_mcp/src/index.js"],
      "env": {
        "SDNEXT_BASE_URL": "http://127.0.0.1:7860",
        "SDNEXT_USERNAME": "user",
        "SDNEXT_PASSWORD": "pass"
      }
    }
  }
}

Or run the Streamable HTTP transport directly:

SDNEXT_MCP_TRANSPORT=http SDNEXT_MCP_PORT=8787 node src/index.js
# → http://127.0.0.1:8787/mcp

Configuration (env vars)

Var Default Meaning
SDNEXT_BASE_URL http://127.0.0.1:7860 WebUI address
SDNEXT_API_PATH /sdapi/v1 v1 prefix
SDNEXT_API_KEY — Bearer token (highest priority)
SDNEXT_USERNAME / SDNEXT_PASSWORD — HTTP Basic auth
SDNEXT_TIMEOUT_MS 300000 Request timeout (generation is slow)
SDNEXT_MCP_TRANSPORT stdio stdio or http
SDNEXT_MCP_HOST / SDNEXT_MCP_PORT 127.0.0.1 / 8787 HTTP transport bind
SDNEXT_MCP_DRY_RUN false 1 = print requests, don't send
SDNEXT_SAVE_DIR — Default dir for save_dir disk writes

Tool families (125 total)

  • Generation — sdnext_txt2img, sdnext_img2img, sdnext_control, sdnext_process_image, sdnext_process_batch, sdnext_detail, sdnext_prompt_enhance
  • Enumerators — sdnext_list_samplers … sdnext_list_masking (29 list tools)
  • Image understanding — sdnext_caption, sdnext_caption_dispatch, sdnext_tagger, sdnext_vqa, sdnext_analyze
  • Models/options — refresh/select/load checkpoint, VAE, UNet, LoRA; sdnext_get_options/sdnext_set_options
  • Server — sdnext_status, sdnext_progress, interrupt/skip, restart/shutdown, log, history, storage, GPU
  • Files/Gallery — sdnext_upload, sdnext_file, delete-, browser/, sdnext_png_info
  • Control processing — sdnext_preprocess, sdnext_mask, sdnext_detect
  • Moderation — sdnext_nudenet, sdnext_prompt_lang, sdnext_image_guard, sdnext_prompt_banned
  • CivitAI — search, model/version, download queue, settings, bookmarks, banned, history (27 tools)
  • Meta — sdnext_list_tools, sdnext_call_api (generic passthrough), autocomplete file mgmt

Run sdnext_list_tools at runtime for the full catalog, or node scripts/dump_tools.mjs.

Typical workflow

sdnext_list_checkpoints   → pick a model
sdnext_list_samplers      → pick a sampler
sdnext_txt2img { prompt, sd_model_checkpoint, sampler_name, width:1024, height:1024, steps:20 }
sdnext_progress           → poll while it runs
sdnext_caption / sdnext_tagger → understand a result

Images are raw base64 PNG in and out (no data: prefix); "upload:<id>" refs are accepted where an image is an input. Pass save_dir on any generation tool to also write the result to disk.

Scripts

npm run check          # syntax-check the entry
node scripts/smoke.mjs # DRY_RUN stdio smoke test (no live server needed)
node scripts/dump_tools.mjs  # export the tool catalog

Layout

See ../framework/ for the architecture documentation (API facts in 01-sdnext-api/, design decisions in 02-mcp-design/). Source under src/: config.js, client.js, image.js, registry.js, index.js, and tools/ (one file per domain).

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选