How to Add Web Search to Your AI Agent

Why your agent needs search

Every LLM has a knowledge cutoff. When your agent needs to answer "What's the current price of Bitcoin?" or "What are the latest React 19 features?", it's working with stale data. Web search fixes this — your agent can look up anything in real time.

Common use cases:

• Answering questions about current events
• Looking up documentation and API references
• Fact-checking claims with real sources
• Researching topics before generating content
• Finding code examples and solutions
• Add Web Search to Cursor IDE via MCP

Method 1: MCP (fastest — 30 seconds)

If you use Claude Code, Claude Desktop, Cursor, or any MCP-compatible tool, this is the fastest way. One config change, zero code.

Step 1: Get a free API key at contextwire.dev (1,000 queries/month, no credit card).

Step 2: Add to your MCP config:

{
  "mcpServers": {
    "contextwire": {
      "url": "https://contextwire.dev/mcp"
    }
  }
}

That's it. Your agent now has 5 search tools:

• ask — "What is X?" → direct answer with sources
• search — web search across 105 engines
• extract — pull clean text from any URL
• research — search academic papers
• batch_search — multiple searches at once
• Add Web Search to Cursor IDE via MCP

Your agent will automatically use these tools when it needs to look something up.

Method 2: Direct API (any language)

Works with any programming language that can make HTTP requests.

Ask a question (get a direct answer)

curl "https://contextwire.dev/api/ask" \
  -H "X-API-Key: YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"q": "What is the current population of Tokyo?"}'

Response:

{
  "answer": "Tokyo's population is approximately 13.96 million...",
  "source": "Wikipedia, WorldPopulationReview",
  "confidence": 0.95,
  "sources": [
    {"title": "Tokyo - Wikipedia", "url": "https://en.wikipedia.org/wiki/Tokyo"},
    {"title": "Tokyo Population 2026", "url": "https://worldpopulationreview.com/..."}
  ]
}

Search (get ranked results)

curl "https://contextwire.dev/api/search" \
  -H "X-API-Key: YOUR_KEY" \
  -d '{"q": "rust async tutorial", "profile": "code", "max_results": 5}'

Python example

import requests

API_KEY = "your-key-here"
BASE = "https://contextwire.dev/api"

def ask(question):
    r = requests.get(f"{BASE}/ask", 
        params={"q": question},
        headers={"X-API-Key": API_KEY})
    return r.json()

def search(query, profile="web"):
    r = requests.get(f"{BASE}/search",
        params={"q": query, "profile": profile, "format": "json"},
        headers={"X-API-Key": API_KEY})
    return r.json()

# Use it
answer = ask("What is the latest version of Python?")
print(answer["answer"])

results = search("FastAPI deployment guide", profile="code")
for r in results["results"][:3]:
    print(f"  {r['title']} — {r['url']}")

Method 3: Node.js SDK

npm install @contextwire/sdk

import { ContextWire } from '@contextwire/sdk';

const cw = new ContextWire('YOUR_API_KEY');

// Ask a question
const answer = await cw.ask('Who won the 2026 Super Bowl?');
console.log(answer.answer);
console.log(answer.sources);

// Search with a specific profile
const results = await cw.search('kubernetes best practices', { 
  profile: 'code', 
  maxResults: 10 
});

// Extract text from a URL
const page = await cw.extract('https://example.com/docs', 'markdown');
console.log(page.content);

// Search academic papers
const papers = await cw.research('transformer architecture improvements');
console.log(papers.papers);

Choosing the right search profile

Profiles determine which of the 105 engines are queried. Using the right profile dramatically improves result quality:

Task	Profile	Why
General questions	web	Google + Bing + DDG + Wikipedia
Code/programming	code	GitHub + StackOverflow + docs
Research papers	academic	arXiv + PubMed + Semantic Scholar
Current events	news	Google News + Bing News
Dev community	devnews	HN + Lobsters + Reddit
Package search	packages	npm + PyPI + crates.io
Security	security	CVE databases

Tips for better search results

1. Be specific. "React useState cleanup" beats "React hooks".
2. Use the right profile. Don't search StackOverflow for news.
3. Use ask for facts, search for exploration. ask synthesizes an answer; search gives you raw results to analyze.
4. Use extract after search. Find the right page, then extract its full content.
5. Use BYOK to save credits. Pass your own OpenRouter key to halve credit usage.