What is Firecrawl /search?
Firecrawl /search is a search API that goes beyond standard web search by returning query-relevant excerpts from each resultânot just generic page descriptions. Instead of forcing your AI agent to read entire pages, Firecrawl uses a trained model to extract the passages that directly answer the userâs question. This approach delivers highly relevant context while using up to 10Ă fewer tokens compared to full-page scraping. On the SimpleQA benchmark, AI agents using Firecrawl /search achieve 94.7% accuracy, outperforming every other provider.
Who it's for
- AI developers building retrieval-augmented generation (RAG) pipelines that need high-quality, token-efficient grounding from the web.
- Agent builders creating autonomous web-surfing agents that must quickly extract answers without wasting compute on irrelevant content.
- Product teams integrating search into conversational interfaces or decision-support tools where speed, accuracy, and cost per query matter.
Key features
Query-aware highlights by default
Every /v2/search call automatically returns highlightsâexcerpts from each page that are most relevant to the query. No extra parameters, output formats, or scrape options are needed. The existing response shape remains unchanged: URL, title, position, and ranking are preserved, and the highlighted text replaces the description (web) or snippet (news).
Markdown-rich excerpts
When page content includes headings, lists, tables, or code, the highlights may contain Markdown formatting. This preserves structure for downstream rendering or processing, making it easy to display structured answers inside chat interfaces or dashboards.
Graceful fallback and per-type control
If Firecrawl cannot generate a highlight for a particular result, the plain website description or snippet is returned insteadâone unavailable page never blocks other results. Developers can also disable highlights by setting highlights: false, reverting to standard descriptions. Image results and Zero Data Retention searches are handled without highlights.
Multi-language SDK and MCP support
Firecrawl /search is accessible via official SDKs (Python, Node.js, cURL, CLI) and the MCP tool firecrawl_search, all returning highlights by default. This makes integration straightforward for teams using different stacks or agent frameworks.
What stands out
âIt outperforms processing full pages while using 10x fewer tokens.â
That headline figureâ94.7% accuracy on SimpleQA with a tenth of the token costâis the productâs core edge. Most search APIs return generic metadata, forcing AI agents to scrape and parse entire pages to find relevant answers. Firecrawl /search eliminates that wasteful step by delivering the needle directly, cutting latency and cost without sacrificing accuracy. The fact that highlights are enabled by default means any existing /v2/search call instantly becomes smarter and cheaper.
Worth checking out ifâŠ
Youâre building AI applications that rely on web-grounding, whether for real-time Q&A, research agents, or automated fact-checking. Firecrawl /search is especially valuable if you want to reduce token consumption without compromising answer quality or if you need search results that are pre-structured for direct use by language models.
