Overview

Exa MCP Server implements a Model Context Protocol (MCP) server that connects AI assistants to Exa’s search and research capabilities. It exposes web search, webpage fetching, code search, and company research tools via a remote MCP endpoint so LLMs and client apps can request structured context (search results, full page content, code snippets) during a conversation or prompt execution.

For developers building or configuring AI assistants, Exa’s MCP server provides a drop-in context source: point your MCP-aware client to the hosted endpoint (https://mcp.exa.ai/mcp) or run the npm package. The server is designed to integrate with desktop assistants, editor extensions, and CLI tools that support the MCP standard, enabling fast, relevant contextual lookups without custom scraping or indexing work.

Features

Fast web search API surfaced over MCP for conversational assistants
Full-page fetch tool to return cleaned page HTML/text for downstream reasoning
Code search across indexed repositories with result snippets ready for model consumption
Company and advanced web research (filterable, domain/date controls) as optional tools
Compatibility with popular MCP clients: Cursor, VS Code MCP, Claude connectors, and more
Simple local invocation via npm package with environment-based API key

Installation / Configuration

Hosted server (recommended)

URL: https://mcp.exa.ai/mcp
Get an Exa API key at https://dashboard.exa.ai/api-keys and provide it to your MCP client as required.

Cursor

// ~/.cursor/mcp.json
{
  "mcpServers": {
    "exa": {
      "url": "https://mcp.exa.ai/mcp"
    }
  }
}

VS Code MCP

// .vscode/mcp.json
{
  "servers": {
    "exa": {
      "type": "http",
      "url": "https://mcp.exa.ai/mcp"
    }
  }
}

Generic mcp-remote (npx)

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

NPM package (run locally)

# install and run the local MCP shim which proxies to Exa with your API key
EXA_API_KEY=your_api_key npx -y exa-mcp-server

Repository and package

GitHub: https://github.com/exa-labs/exa-mcp-server
npm: https://www.npmjs.com/package/exa-mcp-server

Many clients accept the remote URL directly. Others require a small wrapper command (mcp-remote or npx wrapper) — examples above cover the common patterns.

Available Tools

Enabled by default:

Tool	Purpose
web_search_exa	General web search producing clean, model-friendly results
web_fetch_exa	Fetch and return the full content of a specified URL

Off by default (enable via tools parameter in client config):

Tool	Purpose
web_search_advanced_exa	Advanced web search with filters (domains, date ranges, content types)

Deprecated (kept for backward compatibility):

Old tool	Recommended replacement
get_code_context_exa	web_search_exa
company_research_exa	web_search_advanced_exa
crawling_exa	web_fetch_exa
people_search_exa / linkedin_search_exa	web_search_advanced_exa
deep_researcher_*	Exa Research API (separate REST endpoints)

Note: additional tools can be enabled or disabled by passing the tools parameter when configuring your MCP client.

Use Cases

Conversational research assistant: Ask an LLM to summarize the latest news on a topic. The assistant calls web_search_exa for top results, web_fetch_exa to retrieve full articles, then synthesizes a concise summary with sources.
Code discovery during code review: From an editor, query for usages or examples of a function across public repos. web_search_exa returns relevant repository files and snippets so the assistant can explain behavior or suggest refactors.
Due diligence and company research: Use web_search_advanced_exa with domain and date filters to collect press releases, SEC filings, and recent coverage. The assistant aggregates evidence and highlights potential risks or competitors.
Automation in IDEs and chat tools: Configure VS Code or Cursor to provide context-aware answers (e.g., “How do I migrate from X to Y?”). The MCP server supplies real-world migration guides and code samples for the model to reference.

Example flow (high level)

Client sends an MCP tool call to web_search_exa with query “best way to migrate React class components to hooks”.
Server returns ranked results with titles, snippets, and URLs.
Client optionally calls web_fetch_exa on selected URLs to retrieve cleaned article text.
Assistant composes an answer citing the retrieved sources.

Getting Help & Next Steps

Read full API and tool docs: https://docs.exa.ai/reference/exa-mcp
Try the npm package for local development: https://www.npmjs.com/package/exa-mcp-server
Open issues or contribute on GitHub: https://github.com/exa-labs/exa-mcp-server

This server is intended for developers building MCP-compatible integrations. If you need custom tool access or higher throughput, consult Exa’s dashboard and documentation for API keys, rate limits, and enterprise options.

Exa MCP Server: Web, Code, Company Search