Search

Microlink Search turns public search surfaces into structured JSON through the @microlink/google Node.js client. It is designed for SEO tooling, monitoring jobs, research copilots, and agent workflows that need fresh results without maintaining a parser fleet.

This guide stays focused on the package workflow: install the client, choose the right surface with type, then paginate or expand only the results that deserve deeper inspection.

Install and initialize

Install the package, then create one client you can reuse across every supported surface:

npm install @microlink/google

import createGoogleClient from '@microlink/google'

const google = createGoogleClient({
  apiKey: process.env.MICROLINK_API_KEY
})

The only requirement is a Microlink API key. Keep it on the server side or inside your agent backend, not in public browser code.

Search is paid from the first request
PRO

Search does not have a free tier. Reliable public-result collection depends on production-ready managed proxy capacity from the first request, so even small workloads use the same paid delivery model as production workloads.

That point matters in practice:

plan for authenticated server-side usage from day one
do not expose the API key in public clients or shareable tool URLs
expect the same proxy-backed behavior in testing that you will use in production

See

pricing

if you need plan details.

Your first query

Start with the default surface:

const page = await google('technical seo checklist')

console.log(page.results)

When you omit options.type, the client uses search. The same function shape works for every other supported surface.

The mental model

The main API is intentionally small:

const page = await google(query, {
  type: 'search',
  location: 'us',
  period: 'week'
})

Use that model to reason about the package:

query is required and supports standard search operators such as site: or filetype:.
type routes the request to a specific public search surface.
location uses an ISO 3166-1 alpha-2 country code and influences ranking, language, and local intent.
period narrows results to a recent window: hour, day, week, month, or year.

A typical page object starts with a stable core:

{
  "type": "search",
  "results": [
    {
      "title": "Technical SEO Checklist",
      "url": "https://example.com/checklist",
      "description": "A practical checklist for audits and recurring reviews."
    }
  ]
}

Some surfaces also expose extra fields. For example, the default search surface can include knowledgeGraph, peopleAlsoAsk, and relatedSearches.

Choose a surface

Use type to match the workflow to the result shape:

search for broad web retrieval, related questions, and source discovery:
google('technical seo checklist')
news for recent developments, publisher monitoring, and incident tracking:
google('open source llm', { type: 'news' })
images for visual references, image discovery, and asset research:
google('northern lights', { type: 'images' })
videos for video discovery, tutorials, and creator research:
google('cooking tutorial', { type: 'videos' })
places for local entities, addresses, phone numbers, and coordinates:
google('coffee shops denver', { type: 'places' })
maps for richer place metadata such as ratings, hours, and pricing:
google('apple store new york', { type: 'maps' })
shopping for product intelligence, price comparison, and merchant research:
google('macbook pro', { type: 'shopping' })
scholar for papers, citations, and technical or academic research:
google('transformer architecture', { type: 'scholar' })
patents for prior-art research, filing lookup, and invention monitoring:
google('touchscreen gestures apple', { type: 'patents' })
autocomplete for query expansion, demand modeling, and prompt seeding:
google('how to fine tune', { type: 'autocomplete' })

Start with the lightest surface that answers the question. If the workflow is about freshness, switch to news. If it is about product comparison, switch to shopping. If it is about expanding recall, use autocomplete before you fetch full documents.

Localize and filter by time

The quickest way to improve relevance is to tune location and period instead of adding extra parsing logic later:

const page = await google('recetas de pasta', {
  type: 'news',
  location: 'es',
  period: 'week'
})

Use location when the same query means different things across countries. Use period when your workflow depends on recency, such as news monitoring, launch tracking, or alerting.

Paginate naturally

Result pages chain through .next():

const page1 = await google('node.js frameworks')
const page2 = await page1.next()
const page3 = await page2.next()

That makes recurring jobs easier to read than manual cursor handling. A simple loop is often enough:

let page = await google('node.js frameworks')

while (page) {
  for (const result of page.results) {
    console.log(result.title)
  }

  page = await page.next()
}

Expand only the best matches

Search works best as a two-step retrieval model: lightweight results first, deeper content second.

Any result with a url can expose lazy .html() and .markdown() helpers:

const page = await google('site:openai.com function calling guide')

const sources = await Promise.all(
  page.results.slice(0, 3).map(async result => ({
    title: result.title,
    url: result.url,
    html: await result.html(),
    markdown: await result.markdown()
  }))
)

This pattern is especially useful for agents, RAG pipelines, and editorial research. You avoid fetching full documents for every result and only expand the handful that look promising.

Good fit for agents and copilots

@microlink/google is a good tool boundary because the input stays small and the output is already normalized:

Search for the topic with google(query, options).
Inspect the first page of structured results.
Expand a few sources with .html() or .markdown() when the model needs more context.
Call .next() only if confidence is still low or recall needs to improve.

That keeps tool calls cheaper and easier to reason about than handing an agent raw search-result HTML.

Naming and fair-use note

This guide uses the package name @microlink/google and surface names such as search, news, and maps only to identify the public interfaces the client supports.

Microlink Search is an independent Microlink product. It is not affiliated with, endorsed by, or provided by Google. Avoid third-party logos, copyrighted visual assets, or wording that suggests an official partnership when you document or market your own integration.

What's next

Wrap google(query, options) as a tool in your agent framework or internal research workflow.
Add location and period before you build ranking or monitoring logic on top of the results.
Keep .html() and .markdown() for the top matches instead of expanding every result by default.