What it does
Exa MCP Server bridges Claude and other AI assistants to Exa's web search, crawling, and research APIs. It exposes three primary tools: web_search_exa for general web search, web_fetch_exa to retrieve full page content from a known URL, and web_search_advanced_exa for filtered searches with domain, date range, and content-type constraints. The server runs either as a hosted HTTP endpoint (requiring only an API key) or as a local npm package. All searches execute against live web indices, making it suitable for real-time information retrieval.
Who it's for
Developers integrating Claude into research workflows, AI agents that need factual grounding, and engineers building applications where LLMs must access current web data or fetch and analyze specific pages without hallucinating URLs.
Common use cases
- Search the web for recent news or research while working in Claude Code, with results fed directly to the model
- Fetch full page content from a known URL to provide Claude with source material before analysis
- Query code repositories and search specialized domains with date filters and content constraints
- Ground company or people research with filtered web results
- Enable AI agents to autonomously gather and synthesize real-time information for reports or recommendations
Setup pitfalls
EXA_API_KEYis required; retrieve it fromdashboard.exa.ai/api-keysbefore installation- When running locally via npm, the API key must be set as an environment variable in your MCP client config; the hosted HTTP version omits this step
- Many client configurations are available but vary in format — check your specific editor/IDE documentation to ensure correct endpoint URL and parameter names
- If using advanced search features, understand Exa's filtering syntax (domains, date ranges, content types) to avoid overly broad or empty result sets