What it does
The Exa MCP server bridges AI assistants to Exa's search API, enabling real-time web search and content retrieval. By default it exposes two tools: web_search_exa for general web queries and web_fetch_exa to retrieve full page content from a known URL. An optional advanced tool (web_search_advanced_exa) unlocks filtering by domain, date range, content type, and other parameters. All network calls go through Exa's endpoints; the server is stateless and delegates results entirely to the search API.
Who it's for
Backend engineers, researchers, and AI product builders who want Claude or other LLMs to search the live web as part of their workflows—without managing a search index or crawler themselves. Teams using Exa for search infrastructure in production pipelines.
Common use cases
- Fact-check user claims by running web searches and retrieving source content for context
- Build research assistants that gather live market, news, or technical data and synthesize it
- Fetch the full HTML or text of a specific URL to feed large documents into Claude's context window
- Filter web results by domain, date published, or content type (via advanced search) to narrow results for specialized research
- Combine basic and advanced search tools in the same agent—start with broad queries, then refine with filters
Setup pitfalls
- Requires an
EXA_API_KEYenvironment variable if running via npm; omit it if using the hosted server athttps://mcp.exa.ai/mcp - By default, only two tools are enabled;
web_search_advanced_exais off—clients must explicitly opt in to enable it - Deprecated tools (code_context, company_research, crawling, and others) still function for backwards compatibility but are undocumented—migrate to the recommended tools to avoid future breakage