Load URL Content

Fetching a webpage's raw HTML and dumping it into an LLM prompt is a bad idea - you'll burn most of your context window on nav bars, ad markup, and script tags before the model ever gets to the actual content. Load URL Content fetches a page and hands back its readable text instead, which is what makes it usable as an actual knowledge-base input rather than noise.

Inputs and output

url is the page to fetch - default is the placeholder https://example.com, obviously meant to be overwritten, either by you directly or by an upstream node (a search tool's result, for instance) in an agent workflow. is_enable is this pack's standard toggle. The one optional field, with_jina, defaults to on, and it's the interesting part of this node: it routes the fetch through Jina AI's Reader service rather than doing a raw HTTP fetch. Jina's Reader strips a page down to clean, LLM-friendly markdown - handling JavaScript-rendered content, stripping boilerplate, and generally doing the "make this webpage readable" work that a naive HTML fetch can't. Turn with_jina off and you presumably get a more direct fetch of the page instead, which will handle static pages fine but is going to struggle with anything that renders its content client-side via JavaScript.

The output is a single file_content string - the extracted page text, ready to feed into a prompt directly, into this pack's RAG/chunking tools, or into a keyword-search tool that needs a body of text to index.

Where it fits

This is the "give the agent a web page" primitive, and it's the natural pairing for anything that starts with a URL - a link a user pasted, a result from this pack's DuckDuckGo search tool, a GitHub README fetched by URL rather than through the dedicated GitHub tool. Combine it with this pack's Keyword Tool (which turns a block of text into a searchable retrieval tool) and you've got a lightweight way to build a knowledge base out of whatever pages you point it at, without needing a proper crawler or a persistent vector store.

Installing it

Standard pack install - no separate download for this 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

If file_content comes back empty or garbled, the site is the first place to look, not the node - plenty of sites actively block scraping and reader tools (paywalls, aggressive bot detection, login-gated content), and a Jina Reader fetch of a blocked page will come back thin or empty the same way a raw fetch would. There's not much this node can do about a site that doesn't want to be read this way.

If a page you know is readable is coming back poorly formatted or missing content that's clearly rendered client-side, try toggling with_jina - off if Jina's reader is choking on something unusual about the page (some very JS-heavy single-page apps trip up any reader service), on if you're getting raw unusable markup from a direct fetch and need the cleaned-up version instead. It's a one-bit setting, so when the default doesn't work, flipping it is the whole troubleshooting step.