How can an AI agent control ComfyUI?

If you mean an MCP-compatible AI agent operating a live ComfyUI session—not merely explaining a tutorial or generating a one-off script—DCC-MCP can validate API-format graphs against live node contracts, run bounded queue jobs, and retrieve artifacts. DCC-MCP performs work through discoverable typed tools, instance routing, and result validation.

ComfyUI MCP and ComfyUI CLI

The ComfyUI MCP endpoint and ComfyUI CLI workflow share the same DCC-MCP adapter and typed tool catalog. A ComfyUI CLI workflow uses the shared dcc-mcp-cli with --dcc-type comfyui to select the live ComfyUI host; it is not a second, incompatible command line.

dcc-mcp-cli search --query "validate node classes, required inputs, and graph references" --dcc-type comfyui

Can ComfyUI MCP generate free local game assets?

Yes. Adapter source with the comfyui-game-assets Skill supports game UI artwork, icons, transparent PNG assets and GLB props. Image recipes include SD1.5, SDXL, FLUX.2 Klein 4B, Z-Image Turbo and Qwen-Image 2512. BiRefNet removes backgrounds; Hunyuan3D 2 produces untextured shapes; TRELLIS.2 and Pixal3D produce PBR textured meshes.

Have the agent inspect installed models and free VRAM, compare two or three suitable recipes and let the user choose. Reuse an existing choice. Downloads, ComfyUI upgrades and generation should follow the user's authorized scope. For smaller GPUs, compare the SD1.5 draft workflow first; weight size does not guarantee peak memory fit. Local generation remains subject to model and dependency licenses.

If ComfyUI is offline, the agent explains the connection state, offers a concrete startup/configuration or installation plan and waits for authorization. After approval, it completes setup, connection and tool discovery, recipe preflight and the agreed minimal verification. Existing authorization is reused. Installation alone is not asset generation. See the installation and authorization flow.

As of 2026-09-05, these recipes are merged in source; the published 0.1.4 package predates them. Discover and describe the live instance's tools to confirm comfyui-game-assets is present. Node preflight and CPU tests do not establish GPU inference or visual quality. Review alpha edges, lettering, topology, materials and engine import after generation.

• Adapter usage and version guidance
• Local recipe selection, Pixal3D, setup and OOM guidance

What can an AI agent do in ComfyUI?

• validate node classes, required inputs, and graph references.
• submit and wait for one bounded image workflow.
• query terminal status and retrieve a known output artifact.

Capabilities change with the adapter version and loaded Skills. Search and describe tools first instead of guessing current tool names from this page.

Safe operating flow

1. Install and follow the public dcc-mcp Skill.
2. Inspect the existing CLI, adapter, and live hosts; obtain consent before installing software or changing system state.
3. Use health, dcc-types, and list to verify the Gateway and target instance.
4. Search and describe tools for the actual ComfyUI task, then follow every returned next_step.
5. Make one bounded change and verify it through host state, files, previews, logs, or rendered output.

dcc-mcp-cli health
dcc-mcp-cli dcc-types
dcc-mcp-cli list

Copyable prompt

Use the dcc-mcp Skill to connect to my ComfyUI session. Inspect the existing CLI, adapter, and live instance first; ask before installing software or changing system state. Search for and describe typed tools related to "submit and wait for one bounded image workflow", then follow every returned next_step. Do not delete, overwrite, or publish existing work. Make the smallest verifiable change, validate query terminal status and retrieve a known output artifact, and report the instance, tool, result, and evidence path.

Current availability and official source

The current release catalog uses comfyui as the host identifier. Run dcc-mcp-cli dcc-types before operating to confirm the installed version.

• ComfyUI adapter repository: source of truth for installation, compatibility, host-specific capabilities, and troubleshooting.
• All AI + DCC control guides: return to every public application and pipeline integration.
• Agent workflow: shared discovery, call, validation, and diagnostic steps.

This page owns the shared Agent workflow and GEO entry point. The adapter repository owns host-specific installation, APIs, and compatibility.