天气工具(weather_tool)

If you've built anything with function-calling LLMs before, you know the drill: define a JSON schema for the tool, wire it into the model's tools array, write the handler that actually fetches the data, glue the response back into the conversation. Weather Tool collapses all of that into one node. Drop it, plug its output into the tool input on one of LLM Party's agent/LLM nodes, and the model now has a working "what's the weather" function it can call mid-conversation - no separate handler code, no schema you have to hand-write.

This is one of a small family of _tool-suffixed nodes in the pack (Weekday Tool and Work WeChat Tool are its siblings) that all follow the same shape: they don't do anything by themselves. They package themselves into a tool definition and hand it to whichever agent node is doing the actual chat completion. The agent decides, based on what the user asked, whether to call it. That's the mechanism worth understanding before you touch the two fields on the node - this isn't a "run this and get weather" node, it's "make weather available as an option."

The two fields you'll actually set

• city (default 上海市, i.e. Shanghai) - the location this tool reports on. The Chinese-language default is a tell: this pack is built by a Chinese developer with a Chinese-first user base (you'll see the same pattern in Work WeChat and the README's QQ/Feishu integrations), so don't be surprised if things read more naturally with Chinese city names than English ones.
• is_enable - an enum, enable/disable, not a checkbox. Flip it to disable and the tool silently drops out of the agent's toolbox - no error, no warning, the model just stops being able to call it. If your agent stops "knowing" the weather and you didn't unplug anything, this is the first thing to check.

Output is a single tool (STRING) socket. That's not weather data - it's the packaged tool definition, and it only makes sense wired into a node that actually consumes tool inputs. Wiring it into a plain LLM node that doesn't support tool-calling just won't do anything.

Installing it

Same install path as every node in the pack, because it's all one package: search comfyui_LLM_party in ComfyUI Manager and install, or do it by hand -

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

then, inside your ComfyUI Python environment, pip install -r requirements.txt (or, if you're on the portable Windows build, run pip through python_embeded\python.exe). Restart ComfyUI. If you hit weird install errors, the pack ships a requirements_fixed.txt as a fallback dependency set - worth trying before you start hand-debugging version conflicts. This is a genuinely large pack (LLM chat, RAG, MCP tools, local model loaders, social-app integrations all live in the same repo), so expect the same "one custom node drags in a pile of Python deps" friction that any big pack brings - that's a property of the ecosystem, not this node specifically.

Where people get tripped up

Two real gotchas, both structural rather than bug reports:

Wrong slot. The tool output only plugs into nodes built to accept tool definitions. If you're new to this pack and just want an LLM node with weather baked in, you need the tool-calling-capable agent node, not a bare chat node - check what socket type the LLM node you're using actually exposes before you go looking for why the wire won't connect.

The enable/disable enum, not a boolean. Several sibling tool nodes in this pack (Weekday Tool among them) use a plain boolean for their enable switch instead. If you're toggling multiple tool nodes at once and one of them doesn't behave the way you expect, check whether you're actually looking at enable/disable text versus a true/false checkbox - it's an easy thing to fumble when you're copy-pasting a pattern across nodes that look identical at a glance but aren't.

Beyond that, there's no published API key or config requirement visible on the node itself - if it needs one behind the scenes, it isn't something you configure here, so treat this as a plug-and-play tool rather than something to go hunting for credentials for.