Kagi for web search
Web search closes the gap between a model's training cutoff and what the internet says now. Kagi's official server brings a high-quality, ad-free index to that job, and it lands fifth of five here, not because it searches poorly but because its surface is small next to the others.
The case for Kagi is relevance and a clean, commercial-incentive-free index. If your team already pays for that index and trusts its ranking, putting it behind an agent is a natural fit. For broad API coverage or results pre-shaped for LLM consumption, the picks ahead of it do more.
How Kagi fits
The server ships two tools. kagi_search_fetch runs web, news, video, podcast, and image search, accepts domain and date filters and Kagi lenses, and can attach page extracts to results so a single call both searches and reads. kagi_extract fetches a known URL in full and returns it as clean markdown.
That is the whole surface, which is its honest limit: no answer engine, no crawl orchestration, no result objects tuned for a retrieval pipeline. Brave Search offers an independent index with generous query volume. Exa is built for neural, agent-shaped retrieval. Tavily returns results structured for LLM consumption out of a search call, and Perplexity answers questions directly with citations rather than handing back links. Pick Kagi when you specifically want its index quality and the privacy posture, and you are comfortable with two focused tools.
Tools you would use
| Tool | What it does |
|---|---|
| kagi_search_fetch | Runs web, news, video, podcast, and image search with optional page extracts, domain and date filters, and Kagi lenses. |
| kagi_extract | Fetches the full content of a known URL and returns it as clean markdown. |
FAQ
- How many tools does the Kagi MCP server expose for web search?
- Two. kagi_search_fetch runs the search (web, news, video, podcast, image) with domain, date, and lens filters and optional inline page extracts, and kagi_extract pulls a known URL back as clean markdown. There is no separate answer or crawl tool.
- Should I pick Kagi or one of the other web-search servers?
- Pick Kagi when index quality and a privacy-respecting, ad-free search matter most. For high query volume Brave Search fits, for agent-tuned retrieval Exa or Tavily fit, and for direct cited answers Perplexity fits. Kagi ranks fifth here on surface area, not on result quality.