Exa MCP Server

Neural semantic web search and content retrieval MCP server from Exa, optimized for AI agents with clean, structured results.

Data & Enrichment by Exa Labs API Key active

Connect (Remote Server) GitHub Docs npm: exa-mcp-server

Overview

Exa MCP Server connects AI assistants to Exa's neural search engine, exposing web search and full-page fetch tools designed specifically for LLM consumption. Unlike traditional keyword search, Exa uses embeddings-based semantic retrieval and returns clean, parsed content (not raw HTML or SERP snippets), which reduces token waste and improves agent grounding.

The server is officially maintained by Exa Labs and is available two ways: as a remote hosted endpoint at https://mcp.exa.ai/mcp (the recommended setup, used via mcp-remote or as a native Claude Connector) or as a self-hosted npm package. By default it exposes two tools, web_search_exa and web_fetch_exa, with an optional advanced search tool that adds domain, date, and category filters. Several earlier tools (company research, LinkedIn search, code context, crawling) are now deprecated but still functional.

Authentication is via an Exa API key obtained from the Exa dashboard. The hosted server accepts the key as a query parameter (?exaApiKey=...) or an x-api-key header; the npm package reads it from the EXA_API_KEY environment variable. A free tier with rate limits is available without a key.

Tools

Tool	Description
`web_search_exa`	Search the web for any topic and get clean, ready-to-use content optimized for LLM consumption.
`web_fetch_exa`	Get the full content of a specific webpage from a known URL, returned as parsed markdown.
`web_search_advanced_exa`	Advanced web search with full control over filters: category, domain include/exclude, date ranges, subpage crawling, and content options. Disabled by default; enable via tools query parameter.

Setup Guide

Prerequisites

An Exa API key from dashboard.exa.ai/api-keys (a free tier with rate limits is available)
Node.js installed if using the npm package or mcp-remote

Option 1: Remote hosted server (recommended)

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS, %APPDATA%\Claude\claude_desktop_config.json on Windows):

{
  "mcpServers": {
    "exa": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.exa.ai/mcp?exaApiKey=YOUR_EXA_API_KEY"
      ]
    }
  }
}

You can also add Exa as a native Claude Connector via Settings → Connectors → search for Exa.

Option 2: Enable the advanced search tool

Append the tools query parameter to the server URL:

https://mcp.exa.ai/mcp?exaApiKey=YOUR_EXA_API_KEY&tools=web_search_advanced_exa

Option 3: Self-hosted npm package

{
  "mcpServers": {
    "exa": {
      "command": "npx",
      "args": ["-y", "exa-mcp-server"],
      "env": {
        "EXA_API_KEY": "your-exa-api-key"
      }
    }
  }
}

Restart your MCP client after editing the config.

Use Cases

Ground agent answers with up-to-date web information using semantic search rather than keyword matching
Fetch and summarize the full contents of a specific article or documentation page from its URL
Run targeted research queries with date filters and domain include/exclude lists (via web_search_advanced_exa)
Pull cleaned, parsed content directly into a RAG pipeline without writing custom scrapers or HTML parsers
Compare information across multiple sources by searching, then fetching the top results in markdown form

Example Prompts

"Search the web for recent papers on retrieval-augmented generation benchmarks published in the last 6 months."
"Fetch the full content of https://exa.ai/blog and summarize the three most recent posts."
"Find news articles from techcrunch.com about AI agent startups published after January 2026."
"Search for documentation on the MCP specification and quote the section on tool authentication."
"Use advanced search to find blog posts about vector databases, excluding pinecone.io and weaviate.io."

Pros

Officially maintained by Exa Labs, with a hosted endpoint that requires no local install
Returns clean parsed content rather than raw HTML or SERP snippets, reducing token usage
Native Claude Connector support and documented configs for 15+ clients including Cursor, VS Code, Zed, and Warp
Free tier available without an API key for low-volume experimentation

Limitations

Requires an Exa API key for production usage; the free tier is rate-limited
Default tool set is intentionally small (two tools); advanced features and legacy research tools must be explicitly enabled
Several previously available tools (company research, LinkedIn search, code context, crawling) are now deprecated

Alternatives

Tavily MCP: AI-optimized web search API with a similar focus on clean LLM-ready results
Perplexity MCP: Search backed by Perplexity's Sonar models with built-in answer synthesis
Brave Search MCP: Independent web search index with a free API tier