Work WeChat Tool

Compare this node's fields to its sibling, Send to Work WeChat, and you'll spot the tell immediately: there's no content field here. That's not an oversight - it's the whole point. Send to Work WeChat is an action node where you write the message. Work WeChat Tool is a function-calling tool where the LLM writes the message, at conversation time, because you've handed the agent the ability to notify your WeCom group and let it decide when and what to say.

This follows the exact same pattern as Weather Tool and Weekday Tool elsewhere in the pack: the node doesn't do anything on its own. It packages itself into a tool definition and plugs into the tool input of an agent/LLM node that supports function calling. Ask your agent something like "let the team know when this is done" mid-conversation, and if it decides to use the tool, the model itself generates the message content as part of the function call - you never hardcode what gets said.

Setting it up

The webhook side is identical to Send to Work WeChat: add a "群机器人" (group bot) to the WeCom group you want notified, and WeCom hands you back a webhook URL with a key embedded in it. That URL goes in this node's optional url field. As with its action-node sibling, a blank or wrong URL fails silently - nothing will show up in the group and ComfyUI won't tell you why, so that's the first thing to check if the agent claims it sent something and nothing arrived.

Inputs and output

• msgtype - enum, text or markdown (default markdown). WeCom's markdown support is a limited subset, not full markdown - if the model's message renders oddly in the group, try switching this to text to isolate whether it's a formatting quirk or an actual content problem.
• is_enable - boolean; the standard per-tool kill switch. Set it off and this stops being an option the agent can reach for, with no error raised anywhere in the graph.
• url (optional) - the WeCom group-bot webhook URL.

Output is a single tool (STRING) - a packaged function definition, not a message. It only does something useful wired into a node built to consume tool-calling inputs; plugging it into a plain chat-only LLM node won't connect or won't do anything.

Installing it

Same package, same install path as the rest of LLM Party. ComfyUI Manager: search comfyui_LLM_party and install, or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/heshengtao/comfyui_LLM_party

then pip install -r requirements.txt inside your ComfyUI environment, restart.

Common issues

The two failure modes worth knowing are structural, not bugs: wiring the tool output into an LLM node that doesn't actually support tool-calling (check the node's input sockets before assuming the wire should connect), and a webhook url that's blank or copied wrong, which fails without any visible error in the graph - always confirm messages are actually landing in WeCom before assuming your agent's tool-calling logic is at fault. If you want deterministic control over exactly what gets sent and when, rather than letting the model decide, that's what Send to Work WeChat is for instead - the two nodes are deliberately complementary, not interchangeable.