ComfyUI Workflow Analysis

Understand workflow structure and node connections.

MCP Tools

Tool When to use

comfy_read

First step - get human-readable summary of any workflow

comfy_trace

See a specific node's inputs and outputs with slot names

comfy_upstream

Find all nodes feeding INTO a target node

comfy_downstream

Find all nodes FED BY a source node

Workflow

• Summarize: comfy_read(workflow) → pattern, key params, flow

• Deep dive: comfy_trace(workflow, node_id) → specific connections

• Follow data: comfy_upstream /comfy_downstream → trace the graph

What comfy_read Returns

Pattern: Flux txt2img +LoRA

Key Parameters

model: flux-dev.safetensors steps: 20 cfg: 3.5

Variables (3)

$model <- Node 1 (CheckpointLoader)

Flow

[image] 1 -> 5 -> 12 -> 15

What comfy_trace Returns

[Node 577] WanVideoSampler

INPUTS: [0] model <- Node 608 (GetNode, slot 0) [1] image_embeds <- Node 1179 (WanVideoVACEEncode, slot 0)

OUTPUTS: [0] samples -> Node 595 (WanVideoDecode, slot 1)

CLI Equivalents

If MCP isn't available, use CLI commands:

MCP Tool CLI Command

comfy_read

python we_vibin.py analyze workflow.json

comfy_trace

python we_vibin.py trace workflow.json NODE

comfy_upstream

python we_vibin.py upstream workflow.json NODE

comfy_downstream

python we_vibin.py downstream workflow.json NODE

Setup

Requires MCP server. See MCP_SETUP.md.