feat(skills): add comfyui-mcp optional skill for generative image/video workflows#11143
feat(skills): add comfyui-mcp optional skill for generative image/video workflows#11143kshitijk4poor wants to merge 1 commit into
Conversation
|
|
b3bcd24 to
b966368
Compare
|
…eo workflows New optional skill under optional-skills/creative/comfyui-mcp/ with: - SKILL.md: Setup guide (local, cloud, remote), Python helper functions for execute_code, common workflow patterns (txt2img, parameterized gen), queue management, and MCP server integration option. - references/api.md: ComfyUI REST API reference (endpoints, JSON formats). - references/recipes.md: Ready-to-use workflow templates (SDXL txt2img, img2img, Flux). Zero core code changes — skill-only PR. Uses the established skill-as-prompt pattern (like blender-mcp): teaches the agent to interact with ComfyUI's REST API via execute_code.
b966368 to
0f05b18
Compare
|
|
Related to #13271 — both add ComfyUI skills with different approaches (MCP/REST vs CLI-based). |
Summary
New optional skill
comfyui-mcpunderoptional-skills/creative/— teaches Hermes to control a running ComfyUI instance via its REST API.What's Included
execute_code, common workflow patterns (txt2img, parameterized generation), queue management, MCP server integration option, and 10 documented pitfalls.Design
Follows the established skill-as-prompt pattern (like
blender-mcp): the skill teaches the agent to interact with ComfyUI's REST API viaexecute_codeusing a reusable Python helper. No core code changes.Key features of the helper:
COMFYUI_URL/COMFYUI_API_KEYenv vars for flexible deployment (local, cloud, remote)X-API-Keyheader support for Comfy Cloudlist_models()) instead of hardcoded filenames/history/{prompt_id}Testing
Verified end-to-end against a local ComfyUI 0.19.3 instance:
system_stats✓,object_info✓,queue_prompt✓,wait_for_completion✓,get_image✓Zero core code changes — skill-only PR.