Check Web Tool

Most ComfyUI node packs move pixels around. comfyui_LLM_party is one of the few that builds actual LLM agents inside the graph - role-set assistants, multi-tool calling, the works - and Check Web Tool is one of the tools you hand an agent so it can go read a webpage instead of just guessing. It's worth knowing this isn't the only pack doing this kind of thing (comfyui-ollama and the Griptape nodes come up in the same recommendation threads), but LLM party is the one that ships an entire toolbelt like this bundled together rather than making you assemble it from three packs.

How it works

This isn't a "fetch and dump raw HTML" node. It's built for feeding a real webpage into an LLM's context cleanly, with two settings that matter: whether to clean the page up first, and whether to just hand the LLM the whole thing or chunk it for retrieval. The web_url input is optional, which is a tell - in this pack's tool-calling pattern, the agent fills in the URL at call time when it decides "I should check this page," rather than you hardcoding one before the run.

The inputs and outputs that matter

• web_url (optional STRING) - the page to read. Leave it disconnected if you want the LLM to supply it dynamically as part of a tool call.
• with_jina (default true) - routes the fetch through Jina AI's Reader service, which strips a page down to clean text/markdown instead of raw HTML soup. Turn it off only if Jina is unreachable for you or you specifically want the raw page.
• chunk_size / chunk_overlap (200 / 50) - if you're feeding the result into embeddings rather than dumping it straight into the prompt, this is how the text gets split.
• ebd_model (optional EBD_MODEL) - plug in one of the pack's embedding-model loaders here to turn this into a retrieval step instead of a raw-text dump; leave it unconnected for the simple "just fetch the page" behavior.
• base_url / api_key - an OpenAI-compatible endpoint, used when the embedding path needs one.

Output: tool (STRING) - not page content, a tool descriptor. Wire this into the tool/tools input on the pack's LLM or agent node, and the model gets the ability to call it mid-conversation.

How to install it

Search comfyui_LLM_party in ComfyUI Manager, install, restart - or clone it by hand:

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

Then run pip install -r requirements.txt from inside the pack folder, using ComfyUI's own Python interpreter, and restart again. The pack's requirements_fixed.txt is the pinned-version fallback if the main install fights with something else you have installed - this is a large pack (LLM clients, embeddings, OCR, TTS all bundled in one requirements.txt), so version conflicts are more likely here than in a single-purpose node pack.

Common issues & troubleshooting

Fetches come back empty or garbled. That's almost always with_jina doing something you didn't expect - either it's failing silently against a page that blocks the reader service, or you turned it off and are now getting raw HTML tags mixed into what the LLM sees. Toggle it and compare.

The tool never gets called. This node only produces a tool definition - it does nothing on its own. Confirm it's actually wired into the agent/LLM node's tool input, not just sitting connected to nothing, and that your system prompt or agent instructions actually mention it can browse.

API keys live in plain text. Whether you set api_key in config.ini or directly on the node, it ends up either in a config file or baked into the workflow JSON - don't share an exported workflow that still has a live key sitting in a node widget.