comfyui-workflow-editor-mcp
A lightweight MCP server that bridges AI agents with a local ComfyUI instance, enabling them to generate and iteratively refine images, audio, and video through conversational tool calls.
README
comfyui-workflow-editor-mcp
A lightweight MCP (Model Context Protocol) server that bridges AI agents (like Cursor, Claude, etc.) with a local ComfyUI instance. It enables AI agents to generate and iteratively refine images, audio, and video through conversational tool calls — with a graph-based workflow editor for safe, validated modifications.
Features
- Workflow Graph Editor: Safe, validated editing of ComfyUI workflows as directed graphs
- Add, remove, replace, and insert nodes with automatic type-checking and cycle detection
- Search available node types from the ComfyUI
/object_infocatalog - Validate workflows before execution (type mismatches, cycles, orphans, missing inputs)
- Compare workflows, convert between API and frontend formats, generate workflows from templates
Quick Start
Option 1: Via npx (for MCP clients)
No local clone needed. Add to your MCP client configuration (Cursor, Claude, etc.):
"comfyui": {
"command": "npx",
"args": ["-y", "comfyui-workflow-editor-mcp"],
"env": {
"COMFYUI_URL": "http://localhost:8188",
"COMFY_MCP_WORKFLOW_DIR": "/path/to/workflows",
"COMFY_MCP_ASSET_TTL_HOURS": "24"
}
}
Note: ComfyUI must be running at
COMFYUI_URLbefore the MCP client connects.
Option 2: Local development
git clone https://github.com/yar3333/comfyui-workflow-editor-mcp.git
cd comfyui-workflow-editor-mcp
npm install
npm run build
Then start:
| Command | Mode |
|---|---|
npm start |
stdio (for MCP clients) |
npm run dev |
stdio with ts-node |
Configuration
Environment Variables
| Variable | Description | Default |
|---|---|---|
COMFYUI_URL |
ComfyUI base URL | http://localhost:8188 |
COMFY_MCP_WORKFLOW_DIR |
Path to workflow directory | ./workflows |
COMFY_MCP_ASSET_TTL_HOURS |
Asset time-to-live in hours | 24 |
API Tools
Generation Tools
| Tool | Description |
|---|---|
<workflows> |
Available workflows automatically published as tools |
regenerate |
Regenerate a previously generated asset |
Viewing Tools
| Tool | Description |
|---|---|
view_image |
View a generated image inline in chat |
Job Management Tools
| Tool | Description |
|---|---|
get_queue_status |
Get current queue status from ComfyUI |
get_job |
Get job status by prompt_id |
wait_for_job |
Wait for a job to complete with timeout |
list_assets |
List generated assets with optional filtering |
get_asset_metadata |
Get full metadata for a specific asset |
cancel_job |
Cancel a running job by prompt_id |
Configuration Tools
| Tool | Description |
|---|---|
list_checkpoint_models |
List available checkpoint models from ComfyUI |
list_unet_models |
List available UNet models in standard (safetensors) format |
list_unet_gguf_models |
List available UNet models in GGUF format |
Workflow Tools
| Tool | Description |
|---|---|
list_workflows |
List available workflows in the workflow directory |
run_workflow |
Run a specific workflow with parameter overrides |
Workflow Graph Editor Tools
Graph-based tools for safe, validated editing of ComfyUI workflows. All mutations validate type compatibility, detect cycles, and report orphan nodes.
| Tool | Description |
|---|---|
get_workflow_graph |
Get workflow structure as a graph with nodes, links, chains |
search_node_types |
Search available node types from the ComfyUI catalog |
get_node_info |
Get detailed schema for a node type (inputs, outputs, defaults) |
add_node |
Add a node with validation (type-check, cycle detection) |
remove_node |
Remove a node, reporting affected connections and orphans |
connect_nodes |
Create a connection between node outputs and inputs |
disconnect_node_input |
Disconnect an input from a node |
set_node_input |
Change a primitive input value on a node |
insert_node |
Insert a node into an existing link (break-and-reconnect) |
replace_node |
Replace a node type, preserving compatible connections |
validate_workflow |
Full validation: types, cycles, orphans, missing inputs |
find_connection_path |
Find the dependency path between two nodes |
build_basic_workflow |
Generate a workflow from a template (txt2img, img2img, etc.) |
convert_workflow_format |
Convert between ComfyUI API format and frontend JSON format |
diff_workflows |
Compare two workflows, showing added/removed/modified nodes |
Workflow Graph Editor
The workflow graph editor represents ComfyUI workflows as directed graphs, enabling safe modifications with automatic validation.
How It Works
WorkflowGraph— in-memory graph representation of a workflow (nodes, connections, execution order)NodeTypesCatalog— cached catalog of available node types loaded from ComfyUI's/object_infoAPI- Validation rules — every mutation is checked against the same rules ComfyUI uses internally:
| Check | Description |
|---|---|
| Required inputs | All mandatory inputs must be provided |
| Link targets | Links reference existing nodes and valid slot indices |
| Type compatibility | Output types match input type expectations (supports unions and wildcards) |
| Cycle detection | DFS-based cycle detection prevents invalid dependency graphs |
| Orphan detection | Nodes unreachable from any output node are flagged |
| Output node check | At least one output node (SaveImage, SaveAudio, etc.) must exist |
Built-in Templates
build_basic_workflow supports the following templates:
| Template | Description | Nodes |
|---|---|---|
txt2img_basic |
Basic text-to-image | CheckpointLoader, 2×CLIPTextEncode, EmptyLatent, KSampler, VAEDecode, SaveImage |
img2img_basic |
Image-to-image | CheckpointLoader, LoadImage, VAEEncode, 2×CLIPTextEncode, KSampler, VAEDecode, SaveImage |
txt2img_controlnet |
Text-to-image + ControlNet | Basic + ControlNetLoader, LoadImage, ControlNetApply |
txt2img_sdxl |
SDXL text-to-image | CheckpointLoader, CLIPTextEncode (4×), EmptyLatent, KSampler (2×), VAEDecode, SaveImage |
upscale_basic |
Upscale pipeline | CheckpointLoader, ImageUpscaleWithModel, VAEEncode, KSampler, VAEDecode, SaveImage |
Format Conversion
ComfyUI uses two JSON formats:
- API format — flat
{node_id: NodeData}map used by the backend (what the MCP server works with) - Frontend format — structured
{nodes, links, groups}format exported from the ComfyUI UI
Use convert_workflow_format to convert between them.
Workflow System
Workflows are stored as JSON files in the workflows/ directory. The system automatically discovers workflows and exposes them as MCP tools. Parameters are defined using the PARAM_* placeholder system:
PARAM_INT_SEED- Integer parameter for seedPARAM_FLOAT_CFG- Float parameter for CFG scalePARAM_STR_SAMPLER_NAME- String parameter for sampler namePARAM_PROMPT- String parameter for prompt
Test
Prerequisites: ComfyUI running at http://localhost:8188, server built and started.
# Run the test client
npx ts-node test_client.ts
# With custom prompt
npx ts-node test_client.ts -p "a beautiful sunset over mountains"
# Run unit tests
npm test
Project Structure
comfyui-workflow-editor-mcp/
├── src/
│ ├── comfyui_client.ts # HTTP client for ComfyUI API
│ ├── asset_processor.ts # Image processing utilities
│ ├── server.ts # Main entry point
│ ├── models/ # Data models
│ │ ├── asset.ts
│ │ ├── workflow.ts
│ │ ├── workflow_graph.ts # WorkflowGraph — in-memory graph representation
│ │ └── node_types.ts # NodeTypeSchema and related interfaces
│ ├── managers/ # Manager classes
│ │ ├── workflow_manager.ts
│ │ ├── asset_registry.ts
│ │ └── node_types_catalog.ts # NodeTypesCatalog — cached /object_info catalog
│ └── tools/ # MCP tool implementations
│ ├── helpers.ts
│ ├── generation.ts
│ ├── asset.ts
│ ├── job.ts
│ ├── configuration.ts
│ ├── workflow.ts # list_workflows, run_workflow
│ └── workflow_edit.ts # Graph editor tools (add, remove, connect, validate, etc.)
├── workflows/ # Workflow JSON files
├── test_client.ts # Test client
├── package.json
├── tsconfig.json
└── README.md
License
MIT
Author
推荐服务器
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。
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。
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。
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。
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。
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。
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。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
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。