Web Search

Search the live web and get clean, deduplicated, LLM-ready results — titles, URLs and snippets, ranked by how many independent search engines agree on each source.

Powered by our own self-hosted metasearch cluster. No third-party search API sits in the middle, so results are fast, cheap, and not subject to someone else's pricing page.

Endpoints

GET/search?q=next.js+16+release

Web search — up to 30 ranked results.

GET/search?q=what+is+mcp&deep=1

Research mode — also reads the top pages and returns their extracted text.

GET/search?q=india+ai+news&news=1

Fresh results only — last 24 hours.

GET/v1/search?q=...

Versioned alias — same behaviour.

Query Parameters

Parameter	Type	Default	Description
`q`	string	—	✅ Required. The query (2–400 characters)
`limit`	number	10	How many results to return (1–30)
`deep`	0 \| 1	0	Research mode: also read the top pages and return their text
`pages`	number	4	How many pages to read in deep mode (1–8)
`news`	0 \| 1	0	Only results from the last 24 hours

Response

{
  "success": true,
  "query": "next.js 16 release",
  "count": 10,
  "took_ms": 1240,
  "deep": false,
  "sources_read": 0,
  "cached": false,
  "results": [
    {
      "title": "Next.js 16",
      "url": "https://nextjs.org/blog/next-16",
      "snippet": "Next.js 16 is here with Turbopack by default...",
      "engines": ["duckduckgo", "bing"],
      "agreement": 2
    }
  ]
}

In deep mode each result also carries a content field with the readable body of the page (up to ~5,000 characters).

agreement is a trust signal

agreement counts how many independent engines returned that URL. Most results come from a single engine (agreement: 1); roughly 1 in 5 is surfaced by two or more, which is a decent relevance signal. Treat it as a ranking hint, not a guarantee — always rank by your own relevance judgement too.

Feed it straight to an LLM

This is the pattern the service is built for: search, read, then answer with citations.

import requests

r = requests.get(
    "https://api.apimitra.in/search",
    params={"q": "latest ruby release", "deep": 1, "pages": 3, "limit": 6},
    headers={"x-api-key": "YOUR_KEY"},
)

data = r.json()
context = "\n\n".join(
    f"[{i+1}] {x['title']}\nURL: {x['url']}\n{x.get('content', x['snippet'])}"
    for i, x in enumerate(data["results"])
)

# Send `context` as system context to your model and it can cite [1], [2], ...

Examples

# Basic search
curl "https://api.apimitra.in/search?q=best+ai+chat+apps&limit=5" \
  -H "x-api-key: YOUR_KEY"

# Research mode — returns page text too
curl "https://api.apimitra.in/search?q=what+is+mcp&deep=1&pages=3" \
  -H "x-api-key: YOUR_KEY"

# Fresh news only
curl "https://api.apimitra.in/search?q=india+ai+funding&news=1" \
  -H "x-api-key: YOUR_KEY"

Notes

Caching — identical queries are cached for 15 minutes; repeats return in under 1 ms and do not touch the upstream engines.
Deduplication — the same URL from multiple engines is merged into one result, with engines[] listing where it came from.
Noise filtering — known low-value domains are dropped before ranking.
Safety — q is validated and capped at 400 characters; results are returned as data only, never rendered as HTML.

Endpoints​

Query Parameters​

Response​

Feed it straight to an LLM​

Examples​

Notes​