TL;DR Comparison
| API | Free Tier | Credit Card? | Best For |
|---|---|---|---|
| ContextWire | 1,000/mo | No | AI agents, MCP, research |
| Brave Search | 2,000/mo | No | General web search |
| Google CSE | 100/day | Yes (billing account) | Google results specifically |
| SerpAPI | 100/mo | No | SERP scraping, multiple engines |
| Tavily | 1,000/mo (credits) | No | RAG pipelines |
ContextWire — 1,000 Queries/Month
ContextWire is designed for AI agent integration. It searches across 105 engines simultaneously and returns structured, LLM-ready results.
What you get free
- 1,000 queries per month
- Full API access — search, ask, extract, research, batch
- MCP server endpoint (connect any MCP client)
- 22 search profiles (tech, academic, news, etc.)
- BYOK mode: bring your own LLM key for 50% credit savings
- Add Web Search to Cursor IDE via MCP
Quick setup
# Get your key at contextwire.dev, then:
curl "https://contextwire.dev/api/search?q=python+async+best+practices" \
-H "Authorization: Bearer YOUR_KEY"# Python
import requests
resp = requests.get("https://contextwire.dev/api/search", params={
"q": "python async best practices"
}, headers={"Authorization": "Bearer YOUR_KEY"})
results = resp.json()
for r in results["results"]:
print(r["title"], r["url"])Standout features
- Research mode: Multi-step search that follows links, synthesizes info, and returns a cited report
- Extract endpoint: Pull full page content from any URL — useful for reading docs or articles
- 94.3% SimpleQA accuracy: Benchmarked factual accuracy for the ask endpoint
- MCP server: Connect to Claude Code, Cursor, or any MCP client in 30 seconds
- Add Web Search to Cursor IDE via MCP
Brave Search — 2,000 Queries/Month
Brave has its own search index (not a Google wrapper), which means different results — sometimes better for privacy-focused or independent content.
What you get free
- 2,000 queries per month
- Web search with snippets
- No credit card required
- Add Web Search to Cursor IDE via MCP
Quick setup
curl "https://api.search.brave.com/res/v1/web/search?q=python+async" \
-H "X-Subscription-Token: YOUR_KEY"Limitations
- Web search only — no extract, no research mode
- Independent index can miss niche technical content
- No MCP server (though community wrappers exist)
- Add Web Search to Cursor IDE via MCP
Google Custom Search — 100 Queries/Day
Google's official API. You get Google results, but setup is more involved and you need a Google Cloud billing account even for the free tier.
What you get free
- 100 queries per day (~3,000/month)
- Google search results
- Requires billing account (won't charge below limit)
- Add Web Search to Cursor IDE via MCP
Quick setup
curl "https://www.googleapis.com/customsearch/v1?q=python+async&key=YOUR_KEY&cx=YOUR_CX"Limitations
- Must create a Custom Search Engine and configure it for "whole web" search
- Requires Google Cloud project + API key + billing account
- 10 results per query maximum
- $5 per 1,000 queries after free tier
- Add Web Search to Cursor IDE via MCP
SerpAPI — 100 Queries/Month
SerpAPI scrapes real search engine result pages (Google, Bing, Yahoo, etc.) and returns structured JSON. Useful when you need exactly what users see on Google.
What you get free
- 100 queries per month (quite limited)
- Google, Bing, Yahoo, and more
- No credit card for free tier
- Add Web Search to Cursor IDE via MCP
Limitations
- Only 100 queries/month — barely enough for testing
- $50/month for the cheapest paid plan
- SERP scraping — results can break if Google changes layout
- Add Web Search to Cursor IDE via MCP
Others Worth Knowing
Tavily
Popular in the LangChain ecosystem. 1,000 free credits/month, but credits aren't 1:1 with queries (advanced search costs more credits). Good for RAG pipelines.
Exa.ai
Semantic search — you describe what you want in natural language instead of keywords. 1,000 free searches/month. Great for finding similar content or "pages that would answer this question."
DuckDuckGo Instant Answer
Free and unlimited, but limited to instant answers (not full web search). No API key needed. Good for quick factual lookups.
Which One Should You Pick?
| Use case | Best choice | Why |
|---|---|---|
| AI agent / LLM integration | ContextWire | MCP server, research mode, structured output |
| General web search in your app | Brave Search | Highest free quota, clean API |
| Must have Google results | Google CSE | The only way to get actual Google results via API |
| SERP data / SEO tools | SerpAPI | Structured SERP data with rich snippets, PAA, etc. |
| RAG / LangChain pipeline | Tavily or ContextWire | Both offer LLM-optimized output |
| Claude Code / MCP | ContextWire | Only remote MCP search server — 30 second setup |
For most developers building AI-powered tools, ContextWire is the best starting point. It has the most features in the free tier (search + ask + extract + research + batch), the easiest setup (just a URL), and it's the only one with a native MCP server — so it works out of the box with Claude Code, Cursor, and other MCP clients.
Try ContextWire Free
1,000 queries/month. 5 API endpoints. MCP server included. No credit card.
Get Your Free API Key →