NEWResearch Agent: watch an LLM search the web in real time.Try it free →
MCP server is live: give Claude and Cursor the web

Give your agents the web

link.sc is one API to fetch and search anything online. We handle rendering, proxies, and parsing, then return clean markdown and JSON your models can actually use.

200 OK
{
  "url": "https://news.ycombinator.com/",
  "markdown": "# Hacker News\n\n1. Show HN: … (142 points, 63 comments)…",
  "metadata": { "title": "Hacker News", "status": 200 },
  "tokens": 1204
}
POST api.link.sc/v1/fetch · req_01hq8x1 request · 0.42s

TRUSTED BY TEAMS SHIPPING AI IN PRODUCTION

NorthbeamVECTISparallaxQuillioHELIXONMarlowdataspringOKAPI

Ship your first extraction in minutes

Three endpoints cover the whole pipeline, from finding pages to reading them. SDKs for Python and Node, or plain HTTP.

/search

Query the web

One call from question to content. Every result comes back with full-page markdown included.

/fetch

Read any page

Any URL to clean markdown, structured JSON, or screenshots. JS rendering and smart waits included.

/research

Ask a question

An agent searches, reads, and synthesizes a cited report. Every query and fetch streamed live.

Copy
# pip install linksc
from linksc import Link

link = Link(api_key=os.environ["LINK_API_KEY"])

# One call: URL in, clean markdown out
page = link.fetch("news.ycombinator.com")
print(page.markdown[:400])
[ .MD ]response · 0.42s
# Hacker News

1. Show HN: I built a thing
  142 points · 63 comments
2. Ask HN: What are you working on?
  98 points · 210 comments
3. The case for boring technology
  87 points · 41 comments

Fast, reliable, and light on tokens

Web data infrastructure built from the ground up. Rendering, proxies, parsing, and caching are handled for you.

Works on 99.2% of pages

JS-heavy apps, PDFs, and anti-bot protected sites. Automatic rendering, proxy rotation, and smart waits bring back clean data.

link.sc
99%
headless browser
61%
raw HTTP
34%

internal benchmark, 10k mixed URLs, Jun 2026

Speed that feels invisible

p50 0.42s · p95 2.9s

Fetch latency across millions of production requests. Built for real-time agents and live apps.

51ms49ms52ms48ms50ms47ms53ms49ms51ms46ms50msTTFB

time to first byte, sampled from the last 24h

Only the content that matters

92% fewer tokens

Navs, footers, ads, and scripts are stripped. Your model reads content, not markup.

raw HTML38,400 tokens
link.sc markdown3,050 tokens

same page, tokenized with cl100k_base

Plugs into your agent stack

A built-in MCP server exposes fetch and search as tools. One config line and Claude, Cursor, or Windsurf can read the web.

link.sc/mcp/v1MCP2 tools
Claude CodeCursorWindsurfSet it up →

Built for whatever you're building

01

Deep research agents

Let agents read sources end to end instead of working from snippets. Every search result includes full-page markdown.

Learn more →
02

RAG pipelines

Keep knowledge bases fresh with clean, chunk-ready markdown from any site. Semantic structure preserved, noise stripped.

Learn more →
03

AI assistants

Answer with live web context instead of a training-data snapshot. Ground responses in up-to-the-minute information.

Learn more →
04

Lead enrichment

Enrich CRM records with firmographics pulled straight from company sites: team pages, pricing, tech stacks.

Learn more →
05

Competitive intelligence

Monitor competitor sites for pricing changes, new features, and content updates without maintaining scrapers.

Learn more →
06

Agent tooling & MCP

One command gives Claude, Cursor, and Windsurf reliable web tools through our built-in MCP server.

Learn more →

Engineers like it here

SC
Sarah ChenML Engineer, NeuralPath

“We replaced 2,000 lines of Puppeteer code with 5 lines of link.sc fetch calls. Our RAG pipeline went from 60% to 94% retrieval accuracy overnight.”

MR
Marcus RiveraCTO, AgenticHQ

“The search API is a game-changer for our AI agents. Real-time web data grounding reduced hallucinations by over 40% in production.”

PS
Priya SharmaHead of AI, DataForge

“We evaluated every web data API on the market. link.sc was the only one that returned markdown clean enough to feed directly into our embedding pipeline.”

Frequently asked questions

What formats does link.sc return?+

We return clean Markdown by default, optimized for LLM token efficiency. You can also request raw HTML, structured JSON, or screenshots. Our Markdown engine strips ads, navigation, footers, and boilerplate automatically.

How does link.sc handle JavaScript-rendered pages?+

Our Professional and Enterprise plans include full JS rendering via headless browsers. We execute JavaScript, wait for dynamic content to load, and then extract the final rendered DOM. This covers React, Next.js, Vue, Angular, and any SPA framework.

What is your success rate for anti-bot protected sites?+

We maintain a 99.2% success rate across Cloudflare, Akamai, DataDome, and PerimeterX protected sites. Our residential proxy pool rotates through millions of IPs across 195 countries, combined with browser fingerprint rotation.

Can I use link.sc for RAG pipelines?+

Absolutely. link.sc was built specifically for RAG use cases. Our Markdown output is optimized for chunking and embedding. We preserve semantic structure (headings, lists, tables) while removing noise that wastes tokens and degrades retrieval quality.

Do you offer a self-hosted or on-premise option?+

Yes, our Enterprise plan includes an on-premise deployment option. We provide a Docker image that runs on your infrastructure with the same API interface. Contact our sales team for details.

What happens if I exceed my plan limits?+

We never cut off your access mid-request. You'll receive email alerts at 80% and 100% usage. After hitting your limit, additional requests are billed at the overage rate for your plan, or you can upgrade instantly from your dashboard.

Start turning pages into data

500 free requests a month. No card required.