The problem: AI agents are blind

LLMs have a knowledge cutoff. They can reason, write code, and analyze data — but they can't look things up. When your agent needs current stock prices, today's weather, the latest docs for a library, or just a fact it wasn't trained on, it's stuck.

Search APIs solve this. But existing options are expensive ($50-200/month for serious usage), limited in sources, or require complex setup. Most developers end up scraping Google, which breaks constantly.

What is ContextWire?

ContextWire is a search API built specifically for AI agents. It aggregates results from 105 search engines — Google, Bing, DuckDuckGo, Wikipedia, Wolfram Alpha, StackOverflow, GitHub, arXiv, PubMed, and 96 more — into a single, fast API call.

Key features:

  • Free tier — 1,000 queries/month, no credit card
  • 105 engines — not just web search, but code, academic, news, maps, science
  • 22 search profiles — pre-tuned engine combinations for different tasks
  • MCP server — connect to Claude, ChatGPT, or any MCP-compatible agent
  • BYOK — bring your own API key to save credits
  • 94.3% SimpleQA accuracy — independently benchmarked
  • $0 LLM cost — uses Trinity (free via OpenRouter) for answer synthesis
  • Add Web Search to Cursor IDE via MCP

5-minute quickstart

Step 1: Get a free API key at contextwire.dev.

Step 2: Make your first search:

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

That's it. You get back a JSON response with a synthesized answer, sources, and confidence score.

Step 3: For raw search results (no AI synthesis):

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

MCP server setup

If you use Claude Code, Claude Desktop, or any MCP-compatible tool, you can add ContextWire as a tool in one line:

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

This gives your AI agent 5 tools:

  • ask — question → synthesized answer with sources
  • search — query → ranked results from 105 engines
  • extract — URL → clean text extraction
  • research — topic → multi-step research report
  • batch_search — multiple queries in one call
  • Add Web Search to Cursor IDE via MCP

Full MCP setup guide: contextwire.dev/quickstart

22 search profiles

Instead of configuring which engines to use, pick a profile that matches your task:

ProfileBest forEngines
generalGeneral questionsGoogle, Bing, DDG, Wikipedia
codeProgrammingGitHub, StackOverflow, docs
academicResearch papersarXiv, PubMed, Semantic Scholar
newsCurrent eventsGoogle News, Bing News, Reddit
scienceScience questionsWolfram Alpha, Wikipedia, arXiv
legalLegal researchCourt databases, legal sites
financeMarkets/stocksYahoo Finance, market data

Full list of all 22 profiles: contextwire.dev/docs

Bring Your Own Key (BYOK)

ContextWire supports BYOK — pass your own OpenRouter, OpenAI, or Anthropic API key in the request header, and answer synthesis uses your model. This costs only 1 credit instead of 2, so your 1,000 free queries effectively become 1,000 queries instead of 500.

curl "https://contextwire.dev/api/ask" \
  -H "X-API-Key: YOUR_CONTEXTWIRE_KEY" \
  -H "X-LLM-Key: YOUR_OPENROUTER_KEY" \
  -d '{"q": "latest React 19 features"}'

94.3% SimpleQA accuracy

We benchmark ContextWire against SimpleQA — OpenAI's factual accuracy benchmark. ContextWire scores 94.3%, meaning 94 out of 100 factual questions get correct answers with proper sources.

This is achieved through:

  • Multi-engine aggregation (more sources = more reliable)
  • Smart profile routing (each query goes to the most relevant engines)
  • Source verification (answers include citations you can check)
  • Add Web Search to Cursor IDE via MCP

Pricing

PlanPriceQueries/monthNotes
Free$01,000No credit card. BYOK: 1 credit, no key: 2 credits
Pro$10/mo3,500Coming soon
Business$30/mo15,000Coming soon

Sign up free at contextwire.dev.

vs. Tavily, Perplexity, Brave, SerpAPI

A quick comparison with popular alternatives:

FeatureContextWireTavilyPerplexitySerpAPI
Free tier1,000/mo1,000/mo5/day100/mo
Engines1051~51
MCP serverYesYesNoNo
BYOKYesNoNoN/A
SimpleQA94.3%~80%~85%N/A
Search profiles2211N/A

Detailed comparison: contextwire.dev/compare

Try ContextWire free

1,000 searches/month. 105 engines. MCP server included. No credit card.

Get your free API key →