Exa MCP Server – Web and Code Search
Exa's official MCP server gives AI assistants access to web search, code search, and page-reading tools. Use it to retrieve current web context, fetch known URLs as clean Markdown, enable advanced search filters, and ground coding or research workflows in external sources.
Overview
Exa provides an official MCP server that connects AI assistants to Exa's web search, code search, and page-reading capabilities. It is designed for coding agents and research assistants that need current web context, source-backed answers, and clean page content without building a custom search integration.
What the MCP server enables
Exa's Search MCP can be used through the hosted endpoint or the official npm package. The default tools are:
web_search_exa, which searches the web for a topic and returns clean, ready-to-use content.web_fetch_exa, which reads one or more known URLs and returns the page content as clean Markdown.
Exa also documents an optional web_search_advanced_exa tool that can be enabled with a tools URL parameter. This advanced search tool supports more control over category filters, domain restrictions, date ranges, highlights, summaries, and subpage crawling. Older tools such as company research, crawling, people search, LinkedIn search, and deep-search variants remain available for backward compatibility but are documented as deprecated in favor of the newer search and fetch tools.
When to use it
Use Exa MCP when an AI workflow needs current, searchable web or code context. Practical examples include researching recent AI-agent developments, finding Python examples for OAuth 2.0 implementation, fetching and summarizing a known webpage, checking current documentation, comparing public sources, or grounding a coding assistant's answer in fresh external context.
It is especially useful in coding clients such as Cursor, VS Code, Claude Code, Codex, OpenCode, Windsurf, Zed, Gemini CLI, v0, Warp, Kiro, and Roo Code.
Connection and authentication
The official hosted MCP endpoint is:
https://mcp.exa.ai/mcp
For clients that support remote MCP, configure this URL directly using HTTP or Streamable HTTP transport. Exa is also available as a native Claude Connector.
For clients that require a local stdio bridge, Exa documents:
npx -y mcp-remote https://mcp.exa.ai/mcp
For local package execution, Exa documents:
npx -y exa-mcp-server
with EXA_API_KEY set in the environment. The hosted server has a free plan, and Exa recommends adding an API key with the x-api-key header to overcome free-plan rate limits or support production usage.
Key considerations
Use the hosted endpoint for the easiest setup and the npm package or mcp-remote bridge when a client requires local stdio. Keep API keys in environment variables or secure client configuration; never commit them. Optional tools should be enabled deliberately with the tools URL parameter so agents receive only the capabilities they need. Exa documents several older tools as deprecated, so new configurations should prefer web_search_exa, web_fetch_exa, and, when needed, web_search_advanced_exa. Search results and fetched pages should still be reviewed for relevance, freshness, and source quality before being used in high-impact decisions.
Supported Transports
streamable_http
URL: https://mcp.exa.ai/mcp
streamable_http
URL: https://mcp.exa.ai/mcp
stdio
Command: npx
Args:
-ymcp-remotehttps://mcp.exa.ai/mcp
stdio
Command: npx
Args:
-yexa-mcp-server
Frequently Asked Questions
- When should an AI agent use the Exa MCP server?
- Use it when a workflow needs current web or code context, such as researching recent developments, finding implementation examples, fetching a known page, checking documentation, or grounding a coding assistant's answer in external sources.
- What does the Exa MCP server add to an AI agent's capabilities?
- It gives the agent access to Exa's web search, code search, and page-fetching tools so the agent can retrieve current information and clean page content instead of relying only on static model knowledge.
- What can an AI agent access or manage through Exa MCP?
- The agent can run web searches with web_search_exa, fetch known URLs as clean Markdown with web_fetch_exa, and optionally use web_search_advanced_exa for controlled filters, domains, dates, highlights, summaries, and subpage crawling.
- How is authentication configured for the Exa MCP server?
- The hosted server has a free plan. To overcome free-plan rate limits or support production usage, add an Exa API key with the x-api-key header. The local npm package uses the EXA_API_KEY environment variable.
- Which transport should be used for the Exa MCP server?
- Use Streamable HTTP at https://mcp.exa.ai/mcp for direct remote setup. Use npx -y mcp-remote https://mcp.exa.ai/mcp when a client needs a local stdio bridge, or npx -y exa-mcp-server with EXA_API_KEY for the local package.