Exa — Neural Web Search & Research

Exa is a neural search engine. Unlike keyword-based search, it understands meaning — you describe the page you're looking for and it finds it. Returns clean, LLM-ready content with no scraping needed.

MCP server: https://mcp.exa.ai/mcp Free tier: generous rate limits, no key needed for basic tools API key: dashboard.exa.ai/api-keys — unlocks higher limits + all tools Docs: exa.ai/docs GitHub: github.com/exa-labs/exa-mcp-server

Setup

Add the MCP server to your agent config:

# OpenClaw
openclaw mcp add exa --url "https://mcp.exa.ai/mcp"

Or in any MCP config JSON:

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

To unlock all tools and remove rate limits, append your API key:

https://mcp.exa.ai/mcp?exaApiKey=YOUR_EXA_KEY

To enable specific optional tools:

https://mcp.exa.ai/mcp?exaApiKey=YOUR_KEY&tools=web_search_exa,web_search_advanced_exa,people_search_exa,crawling_exa,company_research_exa,get_code_context_exa,deep_researcher_start,deep_researcher_check,deep_search_exa

Tool Reference

Default tools (available without API key)

Optional tools (enable via tools param, need API key for some)

web_search_exa

Fast general search. Describe what you're looking for in natural language.

Parameters:

• query (string, required) — describe the page you want to find
• numResults (int) — number of results, default 10
• type — auto (best quality), fast (lower latency), deep (multi-step reasoning)
• livecrawl — fallback (default) or preferred (always fetch fresh)
• contextMaxCharacters (int) — cap the returned content size

web_search_exa {
  "query": "blog posts about using vector databases for recommendation systems",
  "numResults": 8
}

web_search_exa {
  "query": "latest OpenAI announcements March 2026",
  "numResults": 5,
  "type": "fast"
}

web_search_advanced_exa

The power-user tool. Everything web_search_exa does, plus domain filters, date filters, category targeting, and content extraction modes.

Extra parameters beyond basic search:

Categories:

Examples

Research papers:

web_search_advanced_exa {
  "query": "transformer architecture improvements for long-context windows",
  "category": "research paper",
  "numResults": 15,
  "contents": { "highlights": { "maxCharacters": 3000 } }
}

Company list building with structured extraction:

web_search_advanced_exa {
  "query": "Series A B2B SaaS companies in climate tech founded after 2022",
  "category": "company",
  "numResults": 25,
  "contents": {
    "summary": {
      "query": "company name, what they do, funding stage, location",
      "schema": {
        "type": "object",
        "properties": {
          "name": { "type": "string" },
          "description": { "type": "string" },
          "funding": { "type": "string" },
          "location": { "type": "string" }
        }
      }
    }
  }
}

People search — find candidates with specific profiles:

web_search_advanced_exa {
  "query": "machine learning engineers at fintech startups in NYC with experience in fraud detection",
  "category": "people",
  "numResults": 20,
  "contents": { "highlights": { "maxCharacters": 2000 } }
}

Finding pages similar to a known URL: Use the URL itself as the query — Exa will find semantically similar pages:

web_search_advanced_exa {
  "query": "https://linkedin.com/in/some-candidate-profile",
  "numResults": 15,
  "contents": { "highlights": { "maxCharacters": 2000 } }
}

Recent news with freshness control:

web_search_advanced_exa {
  "query": "AI regulation policy updates",
  "category": "news",
  "maxAgeHours": 72,
  "numResults": 10,
  "contents": { "highlights": { "maxCharacters": 4000 } }
}

Scoped domain search:

web_search_advanced_exa {
  "query": "authentication best practices",
  "includeDomains": ["owasp.org", "auth0.com", "docs.github.com"],
  "numResults": 10,
  "contents": { "text": { "maxCharacters": 5000 } }
}

company_research_exa

One-call company research. Returns business overview, recent news, funding, and competitive landscape.

company_research_exa { "query": "Stripe payments company overview and recent news" }

company_research_exa { "query": "what does Anduril Industries do and who are their competitors" }

people_search_exa

Find professionals by role, company, location, expertise. Returns LinkedIn profiles and bios.

people_search_exa { "query": "VP of Engineering at healthcare startups in San Francisco" }

people_search_exa { "query": "AI researchers specializing in multimodal models" }

get_code_context_exa

Search GitHub repos, Stack Overflow, and documentation for code examples and API usage patterns.

get_code_context_exa { "query": "how to implement rate limiting in Express.js with Redis" }

get_code_context_exa { "query": "Python asyncio connection pooling example with aiohttp" }

crawling_exa

Extract clean content from a specific URL. Handles JavaScript-rendered pages, PDFs, and complex layouts. Returns markdown.

crawling_exa { "url": "https://arxiv.org/abs/2301.07041" }

Good for when you already have the URL and want to read the page.

deep_researcher_start + deep_researcher_check

Long-running async research. Exa's research agent searches, reads, and compiles a detailed report.

Start a research task:

deep_researcher_start {
  "query": "competitive landscape of AI code generation tools in 2026 — key players, pricing, technical approaches, market share"
}

Check status (use the researchId from the start response):

deep_researcher_check { "researchId": "abc123..." }

Poll deep_researcher_check until status is completed. The final response includes the full report.

deep_search_exa

Single-call deep search: expands your query across multiple angles, searches, reads results, and returns a synthesized answer with grounded citations. Requires API key.

deep_search_exa { "query": "what are the leading approaches to multimodal RAG in production systems" }

Supports structured output via outputSchema:

deep_search_exa {
  "query": "top 10 aerospace companies by revenue",
  "type": "deep",
  "outputSchema": {
    "type": "object",
    "required": ["companies"],
    "properties": {
      "companies": {
        "type": "array",
        "items": {
          "type": "object",
          "properties": {
            "name": { "type": "string" },
            "revenue": { "type": "string" },
            "hq": { "type": "string" }
          }
        }
      }
    }
  }
}

Query Craft

Exa is neural — it matches on meaning, not keywords. Write queries like you'd describe the ideal page to a colleague.

Do: "blog post about using embeddings for product recommendations at scale" Don't: "embeddings product recommendations"

Do: "Stripe payments company San Francisco fintech" Don't: "Stripe" (too ambiguous)

• Use category when you know the content type — it makes a big difference.
• For broader coverage, run 2-3 query variations in parallel and deduplicate results.
• For agentic workflows, use highlights instead of full text — it's 10x more token-efficient while keeping the relevant parts.

Token Efficiency

Set maxCharacters on any content mode to control output size.

When to Reach for Which Tool

Docs: exa.ai/docs — Dashboard: dashboard.exa.ai — Support: support@exa.ai

DSH 接入状态（本副本补充）

已配置：Exa MCP server 已写入 DSH 的 MCP 配置（$env:USERPROFILE\.dsh\mcp.json），transport 为 streamable-http：

https://mcp.exa.ai/mcp?exaApiKey=<已配置的 key>

• 生效方式：MCP 由 DSH 的技能与 MCP 管理器插件维护。新增/改动后需要在设置页「Web UI 插件 → 技能与 MCP」确认该 server 处于启用状态；已启用的 server 会真实连接并把工具注册成 mcp__<server>__<tool>。
• 工具名映射：下文写的 web_search_exa、crawling_exa 等在 DSH 里实际是 mcp__exa__web_search_exa、mcp__exa__crawling_exa 这种带前缀的名字。用前先看当前会话可用工具列表，别按裸名调用。
• 若工具没出现：说明 MCP 尚未连接（未启用/未重启），此时先用原生 web_search / web_fetch，并告知用户 MCP 待启用。
• 本技能是工具使用参考（查询写法、参数、category 选择、token 效率），不包含可执行脚本。