What it does
Exa MCP Server exposes Exa's web search and crawling APIs to AI assistants. It provides two tools enabled by default—web_search_exa for general web searches and web_fetch_exa for retrieving full page content—plus web_search_advanced_exa for fine-grained filtering by domain, date, content type, and other parameters. All tools connect via HTTP to Exa's hosted service.
Who it's for
Developers building AI-driven code editors, research tools, or any workflow where Claude or another AI assistant needs live web data. Use it to ground AI responses with current information, fact-check claims during code generation, or retrieve full articles and documentation without manual browsing.
Common use cases
- Fact-check code comments or API documentation claims by searching the web from Claude Code
- Fetch full webpage content for parsing, analyzing, or extracting structured data
- Query with advanced filters (domain, date, content type) from within your IDE
- Ground AI-generated code with current library versions or API docs
- Perform company and product research without switching tools
Setup pitfalls
- Requires a valid Exa API key from
dashboard.exa.ai/api-keys; without it, all tools fail. - When using the npm package locally, set the
EXA_API_KEYenvironment variable; hosted HTTP connectors (Cursor, Claude Desktop) abstract API key handling differently. - The advanced search tool is disabled by default; enable it in your client's MCP config if needed.
- All tools make live network calls to Exa's service—expect network latency and plan for rate limits on high-volume queries.