Exa MCP Server

For comprehensive competitive intelligence, layer multiple query angles using the category:company operator to find competitors, category-specific tools, and recent funding news.

// Layered queries for competitive intelligence
web_search_exa { "query": "category:company companies like [target]", "numResults": 10 }
web_search_exa { "query": "category:company [category] software tools", "numResults": 15 }
web_search_exa { "query": "[category] startup launch funding announcement recently", "numResults": 15 }

To use Exa search capabilities in VS Code, add the following configuration to your .vscode/mcp.json file. You must provide your Exa API key in the EXA_API_KEY environment variable.

{
  "servers": {
    "exa": {
      "command": "npx",
      "args": ["-y", "exa-mcp-server"],
      "env": {
        "EXA_API_KEY": "your_api_key"
      }
    }
  }
}

Exa uses vector embeddings (semantic similarity) rather than keyword matching or boolean logic. To get high-quality results:

• Describe the target page: Instead of searching for a fact (e.g., "X"), describe the type of page you want to find (e.g., "detailed blog post about X written by a practitioner").
• Use natural language: Write queries as grammatical phrases.
• Avoid Boolean operators: Terms like AND, OR, or NOT are treated as regular words, not logical operators. Quotes do not force exact phrase matching.
• Avoid very short queries: 1-2 word queries often produce low-quality, scattered results.
• Handle time semantically: If searching for recent events, calculate the exact dates from your current context and include them in the query (e.g., "published in March 2026") rather than using vague terms like "recent".

Exa is available as a native Claude Connector. You can add it via the UI or manually via the configuration file.

UI Method:

1. Open Claude Desktop Settings (or Customize) and go to Connectors.
2. Search for Exa in the directory.
3. Click + to add it.

Manual Config Method: Add the following to your config file (~/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"]
    }
  }
}

Use web_search_advanced_exa with category: "publication" to find academic papers, arXiv preprints, and scientific research.

Restrictions for category: "publication":

• includeText and excludeText only support single-item arrays. To match multiple terms, include them in the query string or run separate searches.

Supported Parameters:

• query (required)
• numResults
• type ("auto", "fast", "deep", "instant")
• includeDomains / excludeDomains (e.g., ["arxiv.org"])
• startPublishedDate / endPublishedDate (ISO 8601)
• includeText / excludeText (single-item arrays only)
• enableSummary / summaryQuery

Example: Recent Papers

web_search_advanced_exa {
  "query": "transformer attention mechanisms efficiency",
  "category": "publication",
  "startPublishedDate": "2024-01-01",
  "numResults": 15,
  "type": "auto"
}

Use web_search_advanced_exa with category: "people" to find LinkedIn profiles, professional backgrounds, and experts.

Critical Restrictions for category: "people":

• The following parameters cannot be used with this category and will cause errors: startPublishedDate, endPublishedDate, startCrawlDate, endCrawlDate, includeText, excludeText, excludeDomains.
• includeDomains is only supported for LinkedIn domains (e.g., ["linkedin.com"]).
• includeText and excludeText only support single-item arrays.

Example: Discovery

web_search_advanced_exa {
  "query": "VP Engineering AI infrastructure",
  "category": "people",
  "numResults": 20,
  "type": "auto"
}

Use Hard Filters for binary, mechanical criteria that do not require qualitative judgment. These filters should be applied first to eliminate rows efficiently before performing more expensive operations.

Criteria types:

• Date ranges (e.g., "published in 2025")
• Geographic constraints (e.g., "based in SF or NYC")
• Numeric thresholds (e.g., "under $500B market cap")
• Category membership
• Negation filters (e.g., "excluding Novo Nordisk"): Check for the presence of the excluded value and remove matches.

To improve search quality, distinguish between practitioners (who do the work) and commentators (who write about the work). Use these signals to tag your output:

Practitioner signals:

• Shipped products
• Open-source contributions
• Case studies with specific numbers
• First-hand accounts (e.g., "we built X and here's what happened")

Commentator signals:

• Roundup posts
• "Top 10" lists
• Content primarily linking to others' work
• Lack of described first-hand experience

To force an OAuth handshake (allowing users to sign in with their own Exa account), add the login parameter to the URL or use the /mcp/oauth endpoint. This is useful for shared Connectors, Skills, and Plugins.

Using the main endpoint: https://mcp.exa.ai/mcp?login

Using the OAuth endpoint: https://mcp.exa.ai/mcp/oauth

Combining with tools: https://mcp.exa.ai/mcp?tools=web_search_exa,web_fetch_exa,agent_tools&login

To use Exa search capabilities in Cursor, add the following configuration to your ~/.cursor/mcp.json file. You must provide your Exa API key in the EXA_API_KEY environment variable.

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

Use the following command to add the Exa MCP server to Claude Code via stdio transport:

claude mcp add --transport stdio --env EXA_API_KEY=your_api_key exa -- npx -y exa-mcp-server