Answers

POST /answers researches a question for you. It searches the web, reads pages and looks up company records as the task needs, then returns:

json_content: the answer, in the shape you described in json_format
sources: the URLs it read

Facts it couldn’t establish come back null, rather than guessed. Use it when the answer needs searching or several sources. When you already know the page, /scrape with formats=json is cheaper. For a company’s profile from its URL, use /enrich.

Example request

curl -X POST "https://api.prefetch.io/answers" \
  -H "X-API-Key: $PREFETCH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "task": "Who is the CEO of Stripe and in what year was Stripe founded?",
    "json_format": { "ceo": "", "founded_year": 0 },
    "mode": "fast"
  }'

const res = await fetch("https://api.prefetch.io/answers", {
  method: "POST",
  headers: { "X-API-Key": process.env.PREFETCH_API_KEY, "Content-Type": "application/json" },
  body: JSON.stringify({
    task: "Who is the CEO of Stripe and in what year was Stripe founded?",
    json_format: { ceo: "", founded_year: 0 },
    mode: "fast",
  }),
});
const { data } = await res.json();
console.log(data.json_content); // { ceo: "Patrick Collison", founded_year: 2010 }
console.log(data.sources);

import os, requests

r = requests.post(
    "https://api.prefetch.io/answers",
    headers={"X-API-Key": os.environ["PREFETCH_API_KEY"]},
    json={
        "task": "Who is the CEO of Stripe and in what year was Stripe founded?",
        "json_format": {"ceo": "", "founded_year": 0},
        "mode": "fast",
    },
)
data = r.json()["data"]
print(data["json_content"], data["sources"])

Example response

{
  "success": true,
  "data": {
    "json_content": { "ceo": "Patrick Collison", "founded_year": 2010 },
    "sources": [
      "https://stripe.com/",
      "https://stripe.com/newsroom/information",
      "https://stripe.com/newsroom/news/ceo-patrick-collisons-email-to-stripe-employees"
    ],
    "partial": false,
    "key_metadata": { "credits_consumed": 10, "credits_remaining": 69572 }
  },
  "meta": { "requestId": "7c1e2d3f-4a5b-4c6d-8e9f-0a1b2c3d4e5f", "durationMs": 14820 }
}

Parameters

Parameter	Default	Description
`task`	required	The research question, up to 2,000 characters. Name a domain or URL to focus it: named pages are read first.
`json_format`	`{"result": ""}`	An example of the answer object, not a JSON Schema, such as `{"ceo": "", "founded_year": 0}`. Keys and value types are kept, and unknowns come back `null`.
`mode`	`ultra`	`fast`: up to 30 seconds, 10 credits. `ultra`: deeper research across several sources, up to 50 seconds, 100 credits.
`timeout_opts`	none	`{ "milliseconds": 20000, "behavior": "return-partial" }`. Shortens the mode’s deadline (1,000 to 300,000 ms). With `behavior: "return-partial"`, what was found so far comes back with `partial: true`. The other behavior is `fail`.
`tags`	none	Up to 20 labels, 1 to 50 characters each, stored on the request log so you can filter usage.

Over REST the default mode is ultra, at 100 credits. Over MCP, prefetch_answer defaults to fast. Pass mode explicitly if cost matters.

Writing good tasks

Say what you want, not how to find it. “Who is the CEO of Stripe?” works better than “Search Google for Stripe’s CEO.”
Name the site when you know it. “On linear.app, what does the Business plan cost?” reads that site first.
Keep json_format flat and typed. Use 0 for numbers, "" for strings, false for booleans and [] for lists.

Credits and errors

Outcome	Credits
`fast` answer	10
`ultra` answer	100
Failure or timeout	0

When no usable evidence turns up, the request fails with RESEARCH_INCONCLUSIVE and isn’t charged. Rephrase the task, or name the site to research.

Search

Just need the pages? Search and scrape them in one call.

MCP tool

prefetch_answer, for agents.

Get started

Agents and MCP

Company data

Web data

Jobs and monitoring

Platform

Going to production

Example request

Example response

Parameters

Writing good tasks

Credits and errors

Next

Search

MCP tool

​Example request

​Example response

​Parameters

​Writing good tasks

​Credits and errors

​Next

Search

MCP tool

Example request

Example response

Parameters

Writing good tasks

Credits and errors

Next