Exa MCP Server
MCP serversExa's hosted MCP server: web search, page fetch, and advanced filtered search (code, people, companies, papers) for agents.
Install
$ claude mcp add --transport http exa https://mcp.exa.ai/mcp
Update
$ claude mcp add --transport http exa https://mcp.exa.ai/mcp
Hosted HTTP MCP server — updates are server-side, nothing to pull. Re-run add only to change the endpoint (e.g. append ?tools=... to enable more tools). Self-hosted npm users get the latest on the next `npx -y exa-mcp-server`.
When to use it
- You want a coding agent to pull live web results, docs, or real code snippets from GitHub/StackOverflow instead of relying on its training cutoff.
- You need filtered research — companies, LinkedIn people, arXiv papers, SEC filings — via Exa's category and date/domain filters.
- You want to fetch and clean the full text of a known URL for the agent to read.
- You want a hosted, no-key-needed search MCP (or a self-hosted npm one with your own EXA_API_KEY).
Commands & flags
- web_search_exa
- Default tool: search the web on any topic and get clean, ready-to-use page content.
- web_fetch_exa
- Default tool: fetch the full content of a specific webpage from a known URL.
- web_search_advanced_exa
- Off by default: search with full control over domains, dates, categories, and content options.
- ?tools=web_search_exa,web_search_advanced_exa,web_fetch_exa
- Query-param appended to the MCP URL to enable additional/off-by-default tools.
- category: "company" | "people" | "research paper" | "financial report" | "news" | "personal site"
- Scope advanced search to companies, LinkedIn profiles, papers, SEC filings, news, or personal sites.
- claude mcp add --transport http exa "https://mcp.exa.ai/mcp?tools=web_search_advanced_exa"
- Register the hosted server with advanced search enabled (needed for the bundled research skills).
How to use it
This is an MCP server, not a skill or plugin. Register the hosted endpoint once for Claude Code:
claude mcp add --transport http exa https://mcp.exa.ai/mcp
Two default tools are exposed immediately: web_search_exa and web_fetch_exa. To turn on advanced/off-by-default tools, append a tools query param to the URL:
claude mcp add --transport http exa "https://mcp.exa.ai/mcp?tools=web_search_exa,web_search_advanced_exa,web_fetch_exa"
The hosted endpoint needs no key by default; pass ?exaApiKey=YOUR_KEY for higher rate limits. Prefer self-hosting with your own Exa API key? Run the npm package instead:
{
"mcpServers": {
"exa": {
"command": "npx",
"args": ["-y", "exa-mcp-server"],
"env": { "EXA_API_KEY": "your_api_key" }
}
}
}
The README also ships pasteable Claude Skills (company, code, people, financial report, research paper, personal site research) that wrap web_search_advanced_exa with token-isolation and category-filter guidance — paste one into Claude Code and it wires up the MCP and skill for you.
Why it earns its place
Official, first-party MCP server from Exa (the search-for-AI company), published as the exa-mcp-server npm package and listed across Cursor, VS Code, Claude, Codex, Gemini, and more. The hosted HTTP endpoint means zero local setup and no key to manage for the default search/fetch tools, while category filters (company, people, research paper, financial report, news, personal site) and date/domain controls give agents precise, current retrieval well beyond a generic web search. The bundled, copy-paste Claude Skills with built-in token-isolation patterns make it a strong default research backend for Claude Code.
Try it
> Using the Exa MCP, search for recent arXiv papers on transformer attention efficiency published since 2024 and summarize the three most-cited approaches.