DuckDuckGo Tool

Web search is one of the first things anyone wants their LLM agent to have, and it's also one of the more annoying to set up - Google's search API wants a key and a custom search engine ID, Bing's wants an Azure subscription. DuckDuckGo Tool sidesteps all of that. It's a search tool with no API key requirement at all, which is almost certainly why it's got a noticeably better click-through rate than most of the other tool nodes in this pack - "search the web, no signup" is exactly what people are searching for when they land on this page.

What it does

Two required fields, both simple. is_enable is this pack's usual on/off toggle for the tool. searchType is a two-way enum - web for standard search results, image for image search - which tells you this one node covers two different DuckDuckGo search modes rather than needing a separate node per type. Output is a single tool string, wired into an LLM/agent node's tool-list input exactly like this pack's Wikipedia and GitHub tools.

There's no query field on this node itself - unlike Wikipedia Tool, which exposes a query field with a placeholder default, this one has nothing to configure beyond mode and enable state. That fits the tool-calling pattern: the search query itself is generated by the LLM at call time based on what it decides it needs to look up, not something you type in ahead of time on the node.

Why no API key

DuckDuckGo doesn't offer an official public search API the way Google and Bing do - tools that provide "DuckDuckGo search" in Python generally work by querying DuckDuckGo's public-facing search endpoints directly rather than through a documented, stable API contract. That's the tradeoff for skipping the signup step: it's genuinely convenient to set up, but it's leaning on an interface that wasn't built to be a stable API, which is worth knowing before you build something that depends on it running unattended for a long time.

Where it fits

Pair this with this pack's other tools - Wikipedia for reference lookups, GitHub for code and repo questions, this one for general current-events or "what's out there" queries - and give an agent a genuinely broad toolbox without collecting API keys for all three. It's a reasonable default first search tool for anyone just getting a tool-calling agent working in this pack, precisely because there's no credential step blocking you from testing it immediately.

Installing it

Standard pack install, no extra step for this specific node:

• ComfyUI Manager: search "comfyui_LLM_party", install, restart ComfyUI.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/heshengtao/comfyui_LLM_party, then pip install -r requirements.txt inside your ComfyUI Python environment, restart.

Common issues

Because this relies on an unofficial interface rather than a stable, documented API, expect it to be less bulletproof over time than a properly keyed API call would be - search results scraped this way can be rate-limited or occasionally return nothing useful, especially under heavy or automated repeated use, in a way that's harder to diagnose than a clean HTTP error code would be. If web search calls start failing intermittently after working fine earlier, that's a more likely explanation than something wrong in your own workflow.

If you need search results you can depend on for a production pipeline rather than casual/interactive use, this pack's config also supports Google and Bing search (via google_api_key/cse_id or bing_api_key in config.ini) - worth switching to one of those if reliability matters more than avoiding a signup step.