Fix broken JSON from LLMs and scrapers

The problem

LLM output and scraped payloads are constantly almost-JSON: trailing commas, single quotes, unquoted keys, a stray markdown fence. JSON.parse throws, and hand-rolled cleanup breaks on the next edge case you didn't anticipate.

What you get

• Repairs the common failure modes — trailing commas, single quotes, unquoted keys
• Returns valid JSON you can parse directly
• Purpose-built for LLM and scraper output, not just pretty-printing
• One call instead of a growing pile of string replacements

Call it in one request

Plain HTTPS and JSON against /api/v1/json-repair. Use the SDK for typed calls, or hit the endpoint directly — same contract either way.

curl https://zapinner.com/api/v1/json-repair \
  -H "Authorization: Bearer zap_live_••••••" \
  -H "Content-Type: application/json" \
  -d '{"input":"{'a':1, b:2,}"}'

const res = await fetch("https://zapinner.com/api/v1/json-repair", {
  method: "POST",
  headers: {
    Authorization: "Bearer zap_live_••••••",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
  "input": "{'a':1, b:2,}"
}),
})
const data = await res.json()

import { Zapinner } from "@zapinner/sdk"

const zap = new Zapinner({ apiKey: process.env.ZAPINNER_API_KEY })

const result = await zap.jsonRepair({
  "input": "{'a':1, b:2,}"
})

Frequently asked

What kinds of breakage can it fix?

The everyday malformations that make JSON.parse fail — trailing commas, single-quoted strings, unquoted keys, and similar near-miss syntax from model and scraper output.

Is it a good fit for LLM responses?

Yes — it's built for exactly the almost-JSON that language models emit, so you can repair and parse in one step instead of prompting for retries.

What if the input is unrecoverable?

If the payload can't be made into valid JSON, the response says so clearly rather than returning silently-wrong output.

Related solutions