Introducing Highlights: the context that matters

Search the web for AI. Ground every answer.

The web search API for AI agents and LLM apps. Ask in plain language and get ranked, live results with the passages that matter, ready to ground answers and cite sources.

How does hybrid search improve RAG retrieval?

freshness: last_month

Search

queryFanout

hybrid search RAG retrieval

BM25 and vector search accuracy

keyword plus semantic search

  1. [1] docs.example.com / guides / hybrid-search

    high

    Hybrid search: combining keyword and vector retrieval

    “Hybrid search runs a keyword query and a vector query together, then merges the two ranked lists.”
  2. [2] blog.example.com / engineering / bm25-vs-dense

    high

    BM25 vs. dense embeddings for retrieval

    “Keyword scoring catches exact terms that embeddings can miss, such as product codes and names.”
  3. example.org / benchmarks / retrieval

    Retrieval benchmarks for RAG pipelines

    medium

Agent answer

Grounded

Hybrid search pairs keyword scoring with vector similarity and merges the ranked lists [1]. Keyword matching recovers exact terms, like product codes, that embeddings can miss [2].

[1] docs.example.com

[2] blog.example.com

Illustrative agent search with highlights

Backed by Y Combinator. Teams build with Context.dev.

  • Mintlify
  • daily.dev
  • Daydream
  • Super.com
per 10 search results
1 credit
results per query
10–100
free results every month
10,000
results, highlights, pages
1 call
Built for natural language
Send your agent’s question as written. Query fan-out turns it into several searches run in parallel for better recall.
Context, not just links
Every result carries a relevance label. Add query-relevant highlights or full page Markdown in the same call.
Fresh, focused sources
Filter to the last day, week, month, or year. Include or exclude domains and localize results by country.

Add web search to any agent.

Call the SDK from your agent loop, or connect the MCP server to give AI assistants live web search. Each result keeps its title, URL, and relevance together, so your model can cite its sources.

npm install context.dev

search.ts

import ContextDev from 'context.dev';

const client = new ContextDev({ apiKey: 'YOUR_API_KEY' });

const { results } = await client.web.search({
  query: 'How does hybrid search improve RAG retrieval?',
  queryFanout: true,
  freshness: 'last_month',
  highlightsOptions: { enabled: true },
});

const context = results
  .filter((result) => result.relevance === 'high')
  .map((result) => ({
    source: result.url,
    passages: result.highlights.highlights,
  }));

A web search MCP server.

Give Claude Code, Cursor, Codex, or any MCP client live web search. Add one hosted URL and sign in with OAuth, so no API key goes into the config file.

JSON

mcp.json

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

Web search APIs and MCP servers, compared.

What each option returns to your agent, how you connect it as an MCP server, and how usage is priced.

CompareContext.devWeb Search APITavilySearch APIExaNeural search API
What your agent getsRanked results with relevance labels, plus highlights or full-page Markdown in the same callReal-time search, with separate extract, map, and crawl toolsSemantic search with matched excerpts; full content needs a separate scrape
MCP setupOne hosted URL with OAuth sign-in, no API key in the configA remote URL with an API key, or a local server through npxNative Claude connector support
Pricing1 credit per 10 results, 10,000 free results a monthCredits that scale with search depth and extraction optionsMetered per search and per content retrieval

Competitor details as of August 2026. Check each provider for current pricing. Web search APIs and MCP tools for AI agents

From a question to a grounded answer.

  1. Give your agent search

    Add the SDK to your tool loop, or connect the MCP server to your AI assistant.

    client.web.search({ query })
  2. Search with intent

    Pass the question in plain language. Turn on query fan-out and narrow by freshness, domain, or country.

    queryFanout: true
  3. Ground the answer

    Keep high-relevance results, pass their highlights to your model, and cite each URL.

    result.relevance === 'high'

AI web search, answered.

More details in the API documentation.

What is a web search API for AI?

It lets your application or agent search the live web from code and get results it can reason over. Context.dev returns ranked results with titles, URLs, descriptions, and relevance labels, plus optional highlights or full page Markdown in the same call.

How do I give an AI agent web search?

Call client.web.search from your agent’s tool loop with the TypeScript, Python, Ruby, Go, or PHP SDK. To give an AI assistant web search without writing code, connect the Context.dev MCP server.

How much does the web search API cost?

Search costs 1 credit per 10 results. Highlights and Markdown each add 1 credit per 10 results. The free plan includes 1,000 credits every month.

Is there a free web search API?

Yes. The free plan includes 1,000 credits every month with no credit card, enough for 10,000 search results at 1 credit per 10 results.

Is there a web search MCP server?

Yes. Add https://mcp.context.dev/mcp to Claude Code, Cursor, Codex, or another MCP client. Sign-in uses OAuth, so no API key goes into the config file.

Can I get page content for RAG, not just links?

Yes. Enable highlightsOptions for the passages that match the query, or markdownOptions for each full page as Markdown. Check each result’s status code before using its content.

Can I limit search to recent or trusted sources?

Yes. Filter by freshness to the last 24 hours, week, month, or year. Include or exclude domains, localize results by country, and use search operators such as site: and quoted phrases.

How is web search different from Answers?

Search returns relevant pages and optional page content for your application to work with. Answers researches a task and returns structured findings with source URLs.

Give your agents the live web.

Start with 1,000 free credits a month, enough for 10,000 search results.