Exa Semantic Search

Exa provides neural/semantic search via MCP. Use it for high-quality web search that understands context.

When to Use Exa

• Searching for documentation or technical information
• Research requiring semantic understanding
• Finding information where exact keywords are unknown
• Company research and LinkedIn searches
• Deep research tasks

When NOT to Use Exa

• Never use web_search as a substitute – it’s basic keyword matching only
• If Exa fails, troubleshoot Exa – don’t fall back to web_search

Available Tools

The Exa MCP server provides these tools:

• web_search_exa – Semantic web search
• crawling_exa – Crawl and extract web content
• company_research_exa – Research companies
• linkedin_search_exa – Search LinkedIn profiles
• deep_researcher_start – Start deep research task
• deep_researcher_check – Check deep research status

Configuration

Exa is configured as a remote HTTP MCP in ~/.mcp.json:

{
  "exa": {
    "type": "http",
    "url": "https://mcp.exa.ai/mcp?tools=web_search_exa,crawling_exa,company_research_exa,linkedin_search_exa,deep_researcher_start,deep_researcher_check"
  }
}

Usage Examples

Basic Search

Use the Exa MCP tools directly when semantic search is needed.

Deep Research

1. Start with deep_researcher_start for complex topics
2. Poll with deep_researcher_check until complete
3. Get comprehensive, synthesized results

Critical Rules

1. NEVER replace Exa with web_search – they are fundamentally different
2. NEVER use web_search in Task sub-agents as a substitute for Exa
3. If Exa fails, troubleshoot Exa – do not substitute