ComfyUI LLM Party is a Python-based set of nodes and workflows to build LLM-driven agents within ComfyUI, integrating MCP, social apps, local and GGUF/VLM support, and multiple model backends. It provides installation, configuration, and usage guidance across API and local deployment scenarios.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
ComfyUI LLM Party is an LLM Agent Framework that extends ComfyUI with MCP server integration, various social app hooks (Feishu, Discord), and support for local and cloud LLM/VLM backends. It includes GPT-sovits, ChatTTS, GOT-OCR2.0, FLUX prompt nodes, and adapters to OpenAI-like interfaces and multiple model ecosystems (ollama, qwen, gguf, etc.).
How it works
The project exposes nodes for LLM loading, LLM model chaining, and interaction with MCP servers, allowing conversion of server tools into LLM-accessible tools. It supports API-based OpenAI-compatible calls, locally hosted models via Ollama or GGUF/VLM formats, and various model providers through an LLM loader node configured with base_url and api_key. It includes a Knowledge Graph/graphRAG integration pathway and a set of workflows to start different usage modes (start_with_LLM_api, start_with_aisuite, start_with_Ollama, start_with_LLM_local, start_with_LLM_GGUF, start_with_VLM_local, etc.).
Getting started
README outlines multiple installation methods:
- Method 1: Install via comfyui Manager by searching for comfyui_LLM_party and then restart comfyui.
- Method 2: In custom_nodes under ComfyUI root, clone the repo with
git clone https://github.com/heshengtao/comfyui_LLM_party.git. - Method 3: Download zip from CODE, unzip into the
custom_nodessubfolder of the ComfyUI root. Environment deployment steps include runningpip install -r requirements.txtinside the comfyui environment, or usingpath_in_launcher_configuration\python_embeded\python.exe -m pip install -r requirements.txtif using the comfyui launcher. If issues arise, alternatives are inrequirements_fixed.txt. Configuration details include language setting inconfig.ini(zh_CN or en_US), and enabling fast installation viafast_installed(default False). API keys and base_url are configured inconfig.iniper model, with notes for Ollama usage and Google/Bing search tool keys. The documentation also points to example config files and per-model configuration.
Important example usage references are provided as workflow JSON files to load into ComfyUI, including:
workflow/start_with_LLM_api.jsonworkflow/start_with_aisuite.jsonworkflow/ollama.jsonworkflow/start_with_LLM_local.jsonworkflow/start_with_GGUF.jsonworkflow/start_with_VLM_local.jsonworkflow/start_with_VLM_API_for_SD.jsonworkflow/start_with_llava.json(start_with_VLM_GGUF in some contexts)workflow/deepseek-janus-pro.jsonworkflow/qwen-vl.jsonworkflow/图床.jsonfor image hosting
Recent releases
Latest release v0.6.0 (2025-01-15): The convergence of the MCP and OpenAI ecosystems; updates to MCP tool configuration. Prior releases include v0.5.0 (2024-11-23), v0.4.0 (2024-08-03), v0.3.0 (2024-06-30), v0.2.0 (2024-06-03). For v0.6.0, the changelog mentions: The MCP tool has been updated and configuration can be modified in mcp_config.json; reference workflow: workflow/start_with_MCP.json.
Traction
Stars: 2321; Forks: 195; Open issues: 77.
Behind the repo
Not applicable as a startup/company profile in the provided data.
Caveats
License: AGPL-3.0. Language: Python. Created: 2024-04-13. Last push: 2026-07-29. The README notes multiple integration points and environment setup steps, with branch-specific guidance (e.g., only_api branch for API-only usage) and multiple external dependencies via requirements.txt and config.ini.example.