API Function

Not every API call in an agent workflow should be left up to the LLM to decide on. Sometimes you already know exactly which endpoint you want hit and when - you just need the response text flowing into your graph. API Function is that: a plain, synchronous HTTP request node that fires the moment your workflow runs, no LLM judgment involved. It's the "app link" counterpart to api_tool, which wraps the same idea as something the agent chooses to call.

How it works

You give it a URL, and when the graph executes, it makes the request - GET or POST, your choice - and hands back whatever text comes back in the response. There's no decision-making here: this isn't a tool an LLM invokes conditionally, it's a node that runs every time, like any other step in the graph. Use it to pull data in before the LLM sees it (fetch a webpage, hit an internal API, check a status endpoint) or to push data out after the LLM has produced something.

The inputs and outputs that matter

• url - the endpoint being called. No default beyond a placeholder reminding you to fill it in.
• api_key - sent along with the request for endpoints that need authentication.
• parameters - a DICT of whatever the request needs, sent as query params on a GET or as the body on a POST.
• request_type (get / post, default get) and timeout (default 60 seconds) - the two knobs beginners actually need to touch: switch to post if you're submitting data rather than fetching it, and raise timeout if you're hitting something slow.
• One output: text - the raw response body, ready to feed straight into an LLM's prompt, a persona builder, or wherever it needs to go.

How to install it

• ComfyUI Manager - search "comfyui_LLM_party", install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/heshengtao/comfyui_LLM_party, then restart.

Then run pip install -r requirements.txt from inside the comfyui_LLM_party folder, in your ComfyUI Python environment (python_embeded\python.exe -m pip install -r requirements.txt on portable Windows). No special dependency beyond the pack's baseline requirements - this node just wraps a standard HTTP client call.

Common issues & troubleshooting

Request times out. The default 60-second timeout is generous for most APIs but not infinite - if you're calling something genuinely slow (a report-generation endpoint, a cold-starting serverless function), bump the timeout rather than assuming the node is broken.

Response comes back but the LLM can't parse it. text is the raw response - if the API returns JSON, you're getting a JSON string, not parsed fields. Downstream nodes or the LLM itself need to handle that structure; the node doesn't unpack it for you.

Be careful what you paste into api_key and url. This node runs unconditionally with whatever credentials you put on it, saved in your workflow JSON. If you share a workflow file, double-check you haven't left a live API key baked into the node before posting it publicly - custom-node ecosystems in general have a well-documented history of credentials leaking exactly this way, so treat any node that stores a key as something to scrub before sharing.

The pack won't load at all. Separate from this node, there's a real and current install failure mode where comfyui_LLM_party installs via Manager but nodes still refuse to import - usually a CUDA/PyTorch mismatch with no clear supported-version list to check against. Rule that out first if api_function isn't appearing in your node browser.