Exa MCP Server

Exa MCP服务器

将人工智能助手与Exa的搜索功能连接起来：网络搜索、代码搜索和公司研究。

全部文件 | ** | 获取Exa API密钥**

安装

连接到Exa的托管MCP服务器：

https://mcp.exa.ai/mcp

获取API密钥

Cursor

添加 ~/.cursor/mcp.json:

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

VS Code

添加 .vscode/mcp.json:

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

Claude Code

claude mcp add --transport http exa https://mcp.exa.ai/mcp

Claude Desktop

Exa可以作为原生的Claude连接器使用，不需要配置文件或终端命令。

1. 打开克劳德桌面 设置 （或 定制)并前往 连接器
2. 搜索 艾克萨 在目录中
3. 点击 + 添加它

就是这样！克劳德现在可以访问Exa的搜索工具。

Alternative: manual config

添加到配置文件(~/Library/Application Support/Claude/claude_desktop_config.json 在macOS上， %APPDATA%\Claude\claude_desktop_config.json 在Windows上）：

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

Codex

codex mcp add exa --url https://mcp.exa.ai/mcp

OpenCode

添加到您的 opencode.json:

{
  "mcp": {
    "exa": {
      "type": "remote",
      "url": "https://mcp.exa.ai/mcp",
      "enabled": true
    }
  }
}

Antigravity

打开MCP Store面板（从侧面板中的“…”下拉列表中），然后添加一个自定义服务器：

https://mcp.exa.ai/mcp

Windsurf

添加 ~/.codeium/windsurf/mcp_config.json:

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

Zed

添加到Zed设置中：

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

Gemini CLI

添加 ~/.gemini/settings.json:

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

v0 by Vercel

在v0中，选择 提示工具 > 添加MCP 并输入：

https://mcp.exa.ai/mcp

Warp

首选 设置 > MCP服务器 > 添加MCP服务器 并添加：

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

Kiro

添加 ~/.kiro/settings/mcp.json:

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

Roo Code

添加到您的Roo代码MCP配置中：

{
  "mcpServers": {
    "exa": {
      "type": "streamable-http",
      "url": "https://mcp.exa.ai/mcp"
    }
  }
}

Other Clients

对于支持远程MCP的客户端：

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

对于需要mcp-remote的客户端：

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

Via npm Package

将npm包与API密钥一起使用。 获取API密钥.

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

可用工具

默认启用：

默认情况下关闭：

已弃用 （仍然可以向后兼容）：

使用启用其他工具 tools 参数：

https://mcp.exa.ai/mcp?exaApiKey=YOUR_KEY&tools=web_search_exa,web_search_advanced_exa,web_fetch_exa

代理技能（克劳德技能）

Claude Code的现成技能。每项技能都教克劳德如何使用Exa搜索来完成特定任务。复制下拉列表中的内容并将其粘贴到Claude Code中——它将处理其余部分。

Company Research

复制以下内容并将其粘贴到Claude Code中。它将为您设置MCP连接和技能。

Step 1: Install or update Exa MCP

If Exa MCP already exists in your MCP configuration, either uninstall it first and install the new one, or update your existing MCP config with this endpoint. Run this command in your terminal:

claude mcp add --transport http exa "https://mcp.exa.ai/mcp?tools=web_search_advanced_exa"

Step 2: Add this Claude skill

---
name: company-research
description: Company research using Exa search. Finds company info, competitors, news, financials, LinkedIn profiles, builds company lists. Use when researching companies, doing competitor analysis, market research, or building company lists.
context: fork
---

# Company Research

## Tool Restriction (Critical)

ONLY use `web_search_advanced_exa`. Do NOT use `web_search_exa` or any other Exa tools.

## Token Isolation (Critical)

Never run Exa searches in main context. Always spawn Task agents:
- Agent runs Exa search internally
- Agent processes results using LLM intelligence
- Agent returns only distilled output (compact JSON or brief markdown)
- Main context stays clean regardless of search volume

## Dynamic Tuning

No hardcoded numResults. Tune to user intent:
- User says "a few" → 10-20
- User says "comprehensive" → 50-100
- User specifies number → match it
- Ambiguous? Ask: "How many companies would you like?"

## Query Variation

Exa returns different results for different phrasings. For coverage:
- Generate 2-3 query variations
- Run in parallel
- Merge and deduplicate

## Categories

Use appropriate Exa `category` depending on what you need:
- `company` → homepages, rich metadata (headcount, location, funding, revenue)
- `news` → press coverage, announcements
- `people` → LinkedIn profiles (public data)
- No category (`type: "auto"`) → general web results, deep dives, broader context

Start with `category: "company"` for discovery, then use other categories or no category for deeper research.

### Category-Specific Filter Restrictions

When using `category: "company"`, these parameters cause 400 errors:
- `includeDomains` / `excludeDomains`
- `startPublishedDate` / `endPublishedDate`
- `startCrawlDate` / `endCrawlDate`

When searching without a category (or with `news`), domain and date filters work fine.

**Universal restriction:** `includeText` and `excludeText` only support **single-item arrays**. Multi-item arrays cause 400 errors across all categories.

## LinkedIn

Public LinkedIn via Exa: `category: "people"`, no other filters.
Auth-required LinkedIn → use Claude in Chrome browser fallback.

## Browser Fallback

Auto-fallback to Claude in Chrome when:
- Exa returns insufficient results
- Content is auth-gated
- Dynamic pages need JavaScript

## Examples

### Discovery: find companies in a space

web_search_advanced_exa { "query": "AI infrastructure startups San Francisco", "category": "company", "numResults": 20, "type": "auto" }

### Deep dive: research a specific company

web_search_advanced_exa { "query": "Anthropic funding rounds valuation 2024", "type": "deep", "numResults": 10, "includeDomains": ["techcrunch.com", "crunchbase.com", "bloomberg.com"] }

### News coverage

web_search_advanced_exa { "query": "Anthropic AI safety", "category": "news", "numResults": 15, "startPublishedDate": "2024-01-01" }

### LinkedIn profiles

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

## Output Format

Return:
1) Results (structured list; one company per row)
2) Sources (URLs; 1-line relevance each)
3) Notes (uncertainty/conflicts)

Step 3: Ask User to Restart Claude Code

You should ask the user to restart Claude Code to have the config changes take effect.

Code Search

复制以下内容并将其粘贴到Claude Code中。它将为您设置MCP连接和技能。

Step 1: Install or update Exa MCP

If Exa MCP already exists in your MCP configuration, either uninstall it first and install the new one, or update your existing MCP config with this endpoint. Run this command in your terminal:

claude mcp add --transport http exa "https://mcp.exa.ai/mcp?tools=web_search_exa"

Step 2: Add this Claude skill

---
name: code-search-exa
description: Code context using Exa. Finds real snippets and docs from GitHub, StackOverflow, and technical docs. Use when searching for code examples, API syntax, library documentation, or debugging help.
context: fork
---

# Code Context (Exa)

## Tool Restriction (Critical)

ONLY use `web_search_exa`. Do NOT use other Exa tools.

## Token Isolation (Critical)

Never run Exa in main context. Always spawn Task agents:
- Agent calls `web_search_exa`
- Agent extracts the minimum viable snippet(s) + constraints
- Agent deduplicates near-identical results (mirrors, forks, repeated StackOverflow answers) before presenting
- Agent returns copyable snippets + brief explanation
- Main context stays clean regardless of search volume

## When to Use

Use this tool for ANY programming-related request:
- API usage and syntax
- SDK/library examples
- config and setup patterns
- framework "how to" questions
- debugging when you need authoritative snippets

## Query Writing Patterns (High Signal)

To reduce irrelevant results and cross-language noise:
- Always include the **programming language** in the query.
  - Example: use **"Go generics"** instead of just **"generics"**.
- When applicable, also include **framework + version** (e.g., "Next.js 14", "React 19", "Python 3.12").
- Include exact identifiers (function/class names, config keys, error messages) when you have them.

## Output Format (Recommended)

Return:
1) Best minimal working snippet(s) (keep it copy/paste friendly)
2) Notes on version / constraints / gotchas
3) Sources (URLs if present in returned context)

Before presenting:
- Deduplicate similar results and keep only the best representative snippet per approach.

## MCP Configuration

{ "servers": { "exa": { "type": "http", "url": "https://mcp.exa.ai/mcp?tools=web_search_exa" } } }

Step 3: Ask User to Restart Claude Code

You should ask the user to restart Claude Code to have the config changes take effect.

People Search

复制以下内容并将其粘贴到Claude Code中。它将为您设置MCP连接和技能。

Step 1: Install or update Exa MCP

If Exa MCP already exists in your MCP configuration, either uninstall it first and install the new one, or update your existing MCP config with this endpoint. Run this command in your terminal:

claude mcp add --transport http exa "https://mcp.exa.ai/mcp?tools=web_search_advanced_exa"

Step 2: Add this Claude skill

---
name: people-research
description: People research using Exa search. Finds LinkedIn profiles, professional backgrounds, experts, team members, and public bios across the web. Use when searching for people, finding experts, or looking up professional profiles.
context: fork
---

# People Research

## Tool Restriction (Critical)

ONLY use `web_search_advanced_exa`. Do NOT use `web_search_exa` or any other Exa tools.

## Token Isolation (Critical)

Never run Exa searches in main context. Always spawn Task agents:
- Agent runs Exa search internally
- Agent processes results using LLM intelligence
- Agent returns only distilled output (compact JSON or brief markdown)
- Main context stays clean regardless of search volume

## Dynamic Tuning

No hardcoded numResults. Tune to user intent:
- User says "a few" → 10-20
- User says "comprehensive" → 50-100
- User specifies number → match it
- Ambiguous? Ask: "How many profiles would you like?"

## Query Variation

Exa returns different results for different phrasings. For coverage:
- Generate 2-3 query variations
- Run in parallel
- Merge and deduplicate

## Categories

Use appropriate Exa `category` depending on what you need:
- `people` → LinkedIn profiles, public bios (primary for discovery)
- `personal site` → personal blogs, portfolio sites, about pages
- `news` → press mentions, interviews, speaker bios
- No category (`type: "auto"`) → general web results, broader context

Start with `category: "people"` for profile discovery, then use other categories or no category for deeper research on specific individuals.

### Category-Specific Filter Restrictions

When using `category: "people"`, these parameters cause errors:
- `startPublishedDate` / `endPublishedDate`
- `startCrawlDate` / `endCrawlDate`
- `includeText` / `excludeText`
- `excludeDomains`
- `includeDomains` — **LinkedIn domains only** (e.g., "linkedin.com")

When searching without a category, all parameters are available (but `includeText`/`excludeText` still only support single-item arrays).

## LinkedIn

Public LinkedIn via Exa: `category: "people"`, no other filters.
Auth-required LinkedIn → use Claude in Chrome browser fallback.

## Browser Fallback

Auto-fallback to Claude in Chrome when:
- Exa returns insufficient results
- Content is auth-gated
- Dynamic pages need JavaScript

## Examples

### Discovery: find people by role

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

### With query variations

web_search_advanced_exa { "query": "machine learning engineer San Francisco", "category": "people", "additionalQueries": ["ML engineer SF", "AI engineer Bay Area"], "numResults": 25, "type": "deep" }

### Deep dive: research a specific person

web_search_advanced_exa { "query": "Dario Amodei Anthropic CEO background", "type": "auto", "numResults": 15 }

### News mentions

web_search_advanced_exa { "query": "Dario Amodei interview", "category": "news", "numResults": 10, "startPublishedDate": "2024-01-01" }

## Output Format

Return:
1) Results (name, title, company, location if available)
2) Sources (Profile URLs)
3) Notes (profile completeness, verification status)

Step 3: Ask User to Restart Claude Code

You should ask the user to restart Claude Code to have the config changes take effect.

Financial Report Search

复制以下内容并将其粘贴到Claude Code中。它将为您设置MCP连接和技能。

Step 1: Install or update Exa MCP

If Exa MCP already exists in your MCP configuration, either uninstall it first and install the new one, or update your existing MCP config with this endpoint. Run this command in your terminal:

claude mcp add --transport http exa "https://mcp.exa.ai/mcp?tools=web_search_advanced_exa"

Step 2: Add this Claude skill

---
name: web-search-advanced-financial-report
description: Search for financial reports using Exa advanced search. Near-full filter support for finding SEC filings, earnings reports, and financial documents. Use when searching for 10-K filings, quarterly earnings, or annual reports.
context: fork
---

# Web Search Advanced - Financial Report Category

## Tool Restriction (Critical)

ONLY use `web_search_advanced_exa` with `category: "financial report"`. Do NOT use other categories or tools.

## Filter Restrictions (Critical)

The `financial report` category has one known restriction:

- `excludeText` - NOT SUPPORTED (causes 400 error)

## Supported Parameters

### Core
- `query` (required)
- `numResults`
- `type` ("auto", "fast", "deep", "instant")

### Domain filtering
- `includeDomains` (e.g., ["sec.gov", "investor.apple.com"])
- `excludeDomains`

### Date filtering (ISO 8601) - Very useful for financial reports!
- `startPublishedDate` / `endPublishedDate`
- `startCrawlDate` / `endCrawlDate`

### Text filtering
- `includeText` (must contain ALL) - **single-item arrays only**; multi-item causes 400
- ~~`excludeText`~~ - NOT SUPPORTED

### Content extraction
- `textMaxCharacters` / `contextMaxCharacters`
- `enableSummary` / `summaryQuery`
- `enableHighlights` / `highlightsNumSentences` / `highlightsPerUrl` / `highlightsQuery`

### Additional
- `additionalQueries`
- `maxAgeHours` / `livecrawlTimeout`
- `subpages` / `subpageTarget`

## Token Isolation (Critical)

Never run Exa searches in main context. Always spawn Task agents:
- Agent calls `web_search_advanced_exa` with `category: "financial report"`
- Agent merges + deduplicates results before presenting
- Agent returns distilled output (brief markdown or compact JSON)
- Main context stays clean regardless of search volume

## When to Use

Use this category when you need:
- SEC filings (10-K, 10-Q, 8-K, S-1)
- Quarterly earnings reports
- Annual reports
- Investor presentations
- Financial statements

## Examples

SEC filings for a company:

web_search_advanced_exa { "query": "Anthropic SEC filing S-1", "category": "financial report", "numResults": 10, "type": "auto" }

Recent earnings reports:

web_search_advanced_exa { "query": "Q4 2025 earnings report technology", "category": "financial report", "startPublishedDate": "2025-10-01", "numResults": 20, "type": "auto" }

Specific filing type:

web_search_advanced_exa { "query": "10-K annual report AI companies", "category": "financial report", "includeDomains": ["sec.gov"], "startPublishedDate": "2025-01-01", "numResults": 15, "type": "deep" }

Risk factors analysis:

web_search_advanced_exa { "query": "risk factors cybersecurity", "category": "financial report", "includeText": ["cybersecurity"], "numResults": 10, "enableHighlights": true, "highlightsQuery": "What are the main cybersecurity risks?" }

## Output Format

Return:
1) Results (company name, filing type, date, key figures/highlights)
2) Sources (Filing URLs)
3) Notes (reporting period, any restatements, auditor notes)

Step 3: Ask User to Restart Claude Code

You should ask the user to restart Claude Code to have the config changes take effect.

Research Paper Search

复制以下内容并将其粘贴到Claude Code中。它将为您设置MCP连接和技能。

Step 1: Install or update Exa MCP

If Exa MCP already exists in your MCP configuration, either uninstall it first and install the new one, or update your existing MCP config with this endpoint. Run this command in your terminal:

claude mcp add --transport http exa "https://mcp.exa.ai/mcp?tools=web_search_advanced_exa"

Step 2: Add this Claude skill

---
name: web-search-advanced-research-paper
description: Search for research papers and academic content using Exa advanced search. Full filter support including date ranges and text filtering. Use when searching for academic papers, arXiv preprints, or scientific research.
context: fork
---

# Web Search Advanced - Research Paper Category

## Tool Restriction (Critical)

ONLY use `web_search_advanced_exa` with `category: "research paper"`. Do NOT use other categories or tools.

## Full Filter Support

The `research paper` category supports ALL available parameters:

### Core
- `query` (required)
- `numResults`
- `type` ("auto", "fast", "deep", "instant")

### Domain filtering
- `includeDomains` (e.g., ["arxiv.org", "openreview.net"])
- `excludeDomains`

### Date filtering (ISO 8601)
- `startPublishedDate` / `endPublishedDate`
- `startCrawlDate` / `endCrawlDate`

### Text filtering
- `includeText` (must contain ALL)
- `excludeText` (exclude if ANY match)

**Array size restriction:** `includeText` and `excludeText` only support **single-item arrays**. Multi-item arrays (2+ items) cause 400 errors. To match multiple terms, put them in the `query` string or run separate searches.

### Content extraction
- `textMaxCharacters` / `contextMaxCharacters`
- `enableSummary` / `summaryQuery`
- `enableHighlights` / `highlightsNumSentences` / `highlightsPerUrl` / `highlightsQuery`

### Additional
- `userLocation`
- `moderation`
- `additionalQueries`
- `maxAgeHours` / `livecrawlTimeout`
- `subpages` / `subpageTarget`

## Token Isolation (Critical)

Never run Exa searches in main context. Always spawn Task agents:
- Agent calls `web_search_advanced_exa` with `category: "research paper"`
- Agent merges + deduplicates results before presenting
- Agent returns distilled output (brief markdown or compact JSON)
- Main context stays clean regardless of search volume

## When to Use

Use this category when you need:
- Academic papers from arXiv, OpenReview, PubMed, etc.
- Scientific research on specific topics
- Literature reviews with date filtering
- Papers containing specific methodologies or terms

## Examples

Recent papers on a topic:

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

Papers from specific venues:

web_search_advanced_exa { "query": "large language model agents", "category": "research paper", "includeDomains": ["arxiv.org", "openreview.net"], "includeText": ["LLM"], "numResults": 20, "type": "deep" }

## Output Format

Return:
1) Results (structured list with title, authors, date, abstract summary)
2) Sources (URLs with publication venue)
3) Notes (methodology differences, conflicting findings)

Step 3: Ask User to Restart Claude Code

You should ask the user to restart Claude Code to have the config changes take effect.

Personal Site Search

复制以下内容并将其粘贴到Claude Code中。它将为您设置MCP连接和技能。

Step 1: Install or update Exa MCP

If Exa MCP already exists in your MCP configuration, either uninstall it first and install the new one, or update your existing MCP config with this endpoint. Run this command in your terminal:

claude mcp add --transport http exa "https://mcp.exa.ai/mcp?tools=web_search_advanced_exa"

Step 2: Add this Claude skill

---
name: web-search-advanced-personal-site
description: Search personal websites and blogs using Exa advanced search. Full filter support for finding individual perspectives, portfolios, and personal blogs. Use when searching for personal sites, blog posts, or portfolio websites.
context: fork
---

# Web Search Advanced - Personal Site Category

## Tool Restriction (Critical)

ONLY use `web_search_advanced_exa` with `category: "personal site"`. Do NOT use other categories or tools.

## Full Filter Support

The `personal site` category supports ALL available parameters:

### Core
- `query` (required)
- `numResults`
- `type` ("auto", "fast", "deep", "instant")

### Domain filtering
- `includeDomains`
- `excludeDomains` (e.g., exclude Medium if you want independent blogs)

### Date filtering (ISO 8601)
- `startPublishedDate` / `endPublishedDate`
- `startCrawlDate` / `endCrawlDate`

### Text filtering
- `includeText` (must contain ALL)
- `excludeText` (exclude if ANY match)

**Array size restriction:** `includeText` and `excludeText` only support **single-item arrays**. Multi-item arrays (2+ items) cause 400 errors. To match multiple terms, put them in the `query` string or run separate searches.

### Content extraction
- `textMaxCharacters` / `contextMaxCharacters`
- `enableSummary` / `summaryQuery`
- `enableHighlights` / `highlightsNumSentences` / `highlightsPerUrl` / `highlightsQuery`

### Additional
- `additionalQueries`
- `maxAgeHours` / `livecrawlTimeout`
- `subpages` / `subpageTarget` - useful for exploring portfolio sites

## Token Isolation (Critical)

Never run Exa searches in main context. Always spawn Task agents:
- Agent calls `web_search_advanced_exa` with `category: "personal site"`
- Agent merges + deduplicates results before presenting
- Agent returns distilled output (brief markdown or compact JSON)
- Main context stays clean regardless of search volume

## When to Use

Use this category when you need:
- Individual expert opinions and experiences
- Personal blog posts on technical topics
- Portfolio websites
- Independent analysis (not corporate content)
- Deep dives and tutorials from practitioners

## Examples

Technical blog posts:

web_search_advanced_exa { "query": "building production LLM applications lessons learned", "category": "personal site", "numResults": 15, "type": "deep", "enableSummary": true }

Recent posts on a topic:

web_search_advanced_exa { "query": "Rust async runtime comparison", "category": "personal site", "startPublishedDate": "2025-01-01", "numResults": 10, "type": "auto" }

Exclude aggregators:

web_search_advanced_exa { "query": "startup founder lessons", "category": "personal site", "excludeDomains": ["medium.com", "substack.com"], "numResults": 15, "type": "auto" }

## Output Format

Return:
1) Results (title, author/site name, date, key insights)
2) Sources (URLs)
3) Notes (author expertise, potential biases, depth of coverage)

Step 3: Ask User to Restart Claude Code

You should ask the user to restart Claude Code to have the config changes take effect.

链接

• 文档
• 获取Exa API密钥

建于❤️ 通过Exa