JSON Parse

The moment your LLM starts returning structured output, you need a way to crack the JSON open in the graph. JSON Parse is that tool: it takes a JSON string - the kind a Chat node hands back when you ask for structured output - and extracts up to eight values by dot/bracket paths, each on its own output port. No Python, no scripting node, just a path string per port.

How it works

You give it json_text plus up to eight path fields. Paths are the familiar dot/bracket syntax: positive, params.width, items[0].name, items[0]["name"]. It walks the parsed object and lands each value on the matching out1–out8 port as a STRING (numbers and booleans get stringified, nested objects come back as JSON text). There's also a obj output that carries the whole parsed object on the pack's SIMPLECHAT_JSON type, for chaining into other SimpleChat nodes.

Three niceties baked in:

• Code fences auto-stripped - strip_code_fence defaults on, so a reply wrapped in ```json ... ``` still parses. LLMs love wrapping their JSON; this saves you a cleanup step.
• A default value - the default input (default "") is what a port gets when a path isn't found. Keeps missing keys from erroring the whole graph.
• Strict-ish parsing - it's a straightforward json.loads with fence-stripping, so well-formed output is fine and malformed output errors loudly (unlike the forgiving parsers in Prompt JSON Unpack).

Why it beats manual copy-paste

Because it runs live. In the flagship loop, Chat → JSON reply → this node → eight string ports wired into your sampler's prompt and parameter slots. Ask the LLM for {"positive": ..., "negative": ..., "seed": ...} and re-run with a different request to iterate without ever touching the node. It's the manual version of Prompt JSON Unpack - which is exactly the trade: this one takes any schema via paths, the other is fixed-schema but gives you typed INT/FLOAT/SAMPLER outputs. If your JSON matches the Anima prompt schema, use Prompt JSON Unpack and skip the path typing; if your schema is your own, this is the one.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/Moeblack/ComfyUI-SimpleChat

Restart, or use Manager and search "ComfyUI-SimpleChat". Only dependency: aiohttp.

Gotchas

Eight ports is a hard ceiling - for sixteen, use the pack's JSON Parse (16) variant. Values come out as strings, so if you wire out3 (a number) into an INT input you'll need Prompt JSON Unpack-style casting or a convert node; this node won't do it. And remember it needs a string of JSON - if your upstream hands you the already-parsed SIMPLECHAT_JSON object, you want JSON -> Vars or a direct obj consumer instead.