Search MCP Tools Compared

Three jobs a search tool can do

A search tool returns a ranked list of results, usually with titles, URLs and snippets. Use it when the agent should choose its own sources. A fetch or extract tool reads known URLs and returns clean text, as Exa's web_fetch_exa and Tavily's tavily_extract do. Full pages can be long, so limit how many you pull at once.

An ask or research tool returns one answer, or a report, built from several sources. Hyperresearch's ask returns a short answer with quotes matched word for word against their pages, or a plain not-found message. Tavily's tavily_research and Perplexity's research tool take a broader approach that returns a longer synthesized result.

Some tools sit between these jobs. Brave's brave_llm_context returns pre-extracted content meant for grounding, and Firecrawl's scrape tool returns the content of a URL you name. Both return more text than a ranked list, and both leave the decision about what to trust to the agent.

Tool lists from the vendor docs

Hyperresearch Answers exposes ask and search. Search returns ranked sources with matching passages without summarizing them. Both draw on the open web and about 90 structured sources.

Exa's web_search_exa and web_fetch_exa are on by default. Its web_search_advanced_exa tool is opt-in for filters and date ranges. Tavily exposes tavily_search, tavily_extract, tavily_crawl, tavily_map and tavily_research. Perplexity exposes perplexity_search, perplexity_ask, perplexity_research and perplexity_reason. Brave exposes brave_web_search, brave_news_search, brave_local_search and several others, plus brave_llm_context for grounding.

In its keyless tier, Firecrawl names Search, Scrape and Parse, and its documentation points to a separate tools page for the full list.

Read the list before you enable it

The model reads each tool's name and description when it decides what to call. A server with many tools uses more context on every turn, so disable the ones you will not use. The MCP tools specification says clients should treat tool annotations as untrusted unless they come from a trusted server.

Compare parameters too. Brave's web search takes a count and a freshness filter, while Exa's advanced tool adds filters, date ranges and geo targeting. Test each tool with one query before you rely on it, and note how large its output is.

Checklist

• List the tools a server exposes using your client's MCP panel.
• Sort each tool into a search, fetch or answer job.
• Enable one search tool and one fetch tool to start.
• Add an answer tool only when you want cited, short responses.
• Re-check the list after each vendor update, because tool names can change.

Questions & answers

Sources for this entry

• Exa MCP server reference
• Tavily MCP documentation
• Perplexity MCP server documentation
• Brave Search MCP server (GitHub)
• MCP specification: Tools

Vendor descriptions are attributed to the vendor. Prices change; confirm with each vendor.

Continue this reading path

• Servers
• Search MCP Server
• Resource directory

Cite this entry