Google Search Tool

This one doesn't do anything by itself - it's a tool definition, not an action. What it produces is a tool output that you plug into the tools input of one of the pack's LLM nodes, and from that point on the model can decide, mid-conversation, to actually go search Google instead of guessing from training data. That's the whole point of comfyui_LLM_party's tool-calling layer: chat nodes on their own can't browse, but wire enough of these tool nodes into tools and your agent starts looking less like a static chatbot and more like something that can go fetch a current answer.

The two required fields are simple: is_enable just turns the tool on or off without deleting the node (handy for A/B-ing a workflow with and without search), and searchType picks between web and image results - the latter lets the model pull back image search hits rather than text snippets, useful if you're building something that needs to find reference pictures rather than facts.

Where it gets slightly more involved is authentication. Google Custom Search isn't free-for-all - you need a Google API key and a Custom Search Engine (CSE) ID, both of which are optional fields on this node because you're expected to configure them once centrally instead. The pack's config.ini has google_api_key and cse_id fields precisely for this: fill them in there and every google_tool node in every workflow just works without you typing credentials into the graph itself (and without them ending up saved inside a workflow JSON you might share with someone). If you'd rather keep it node-local - say you're testing with a throwaway key - you can fill google_api_key and google_CSE_ID directly on the node instead. Just don't do both and forget which one you set; an empty search result with no error is the usual symptom of a key that's configured in one place but expected in the other.

Installing it is the same as installing the whole pack, since this ships as part of comfyui_LLM_party rather than standalone: search "comfyui_LLM_party" in ComfyUI Manager and install, or git clone https://github.com/heshengtao/comfyui_LLM_party into custom_nodes and restart. After that, run pip install -r requirements.txt from inside the pack's folder, using ComfyUI's own Python (portable installs need the explicit python_embeded\python.exe -m pip install -r requirements.txt form - using your system pip is the most common way people end up with half-installed dependencies).

Where this actually breaks in practice: the model has to be told, in its system prompt or via is_tools_in_sys_prompt on the LLM node, that it has a search tool available and when to use it - not every model calls tools reliably out of the box, and a model with no native function-calling support will mostly just ignore the tool unless you explicitly nudge it in the prompt. Google's Custom Search API also has a free-tier daily query cap, so if search silently stops returning results partway through a long session, check your quota before you check your code. And because this is fundamentally a Google Custom Search wrapper, not a general web crawler, it returns whatever a CSE query returns - no page-content fetching beyond the snippet, so don't expect the model to read a full article through this tool alone.