exa.ai-websearch-api（EXA AI websearch API 搜索）

Exa API Setup Guide

Exa | Web Search API, AI Search Engine, & Website Crawler

Your Configuration

Keywords

ai search engine,serp api,best ai search engine,search api,deep research api,web search api,web search ai,ai search api,deepresearch api,web crawling api,website crawler,metaphor,exa,api,search,ai,llms

API Key Setup

Environment Variable

export EXA_API_KEY="YOUR_API_KEY"

.env File

EXA_API_KEY=YOUR_API_KEY

🔌 Exa MCP Server

Give your AI coding assistant real-time web search with Exa MCP.

Remote MCP URL:

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

Available tools: web_search_exa, get_code_context_exa, company_research_exa, crawling_exa, linkedin_search_exa, deep_researcher_start

HTTP config (Cursor, Claude Code, Codex):

{
  "mcpServers": {
    "exa": {
      "type": "http",
      "url": "https://mcp.exa.ai/mcp?exaApiKey=YOUR_API_KEY",
      "headers": {}
    }
  }
}

Local install (Claude Desktop):

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

📖 Full docs: docs.exa.ai/reference/exa-mcp

Quick Start (cURL)

cURL

curl -X POST 'https://api.exa.ai/search' \
  -H 'x-api-key: YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
  "query": "React hooks best practices 2024",
  "type": "auto",
  "num_results": 10,
  "contents": {
    "text": {
      "max_characters": 20000
    }
  }
}'

Function Calling / Tool Use

Function calling (also known as tool use) allows your AI agent to dynamically decide when to search the web based on the conversation context. Instead of searching on every request, the LLM intelligently determines when real-time information would improve its response—making your agent more efficient and accurate.

Why use function calling with Exa?

• Your agent can ground responses in current, factual information
• Reduces hallucinations by fetching real sources when needed
• Enables multi-step reasoning where the agent searches, analyzes, and responds

📚 Full documentation: https://docs.exa.ai/reference/openai-tool-calling

OpenAI Function Calling

import json
from openai import OpenAI
from exa_py import Exa

openai = OpenAI()
exa = Exa()

tools = [{
    "type": "function",
    "function": {
        "name": "exa_search",
        "description": "Search the web for current information.",
        "parameters": {
            "type": "object",
            "properties": {"query": {"type": "string", "description": "Search query"}},
            "required": ["query"]
        }
    }
}]

def exa_search(query: str) -> str:
    results = exa.search(query=query, type="auto", num_results=10, contents={"text": {"max_characters": 20000}})
    return "\n".join([f"{r.title}: {r.url}" for r in results.results])

messages = [{"role": "user", "content": "What's the latest in AI safety?"}]
response = openai.chat.completions.create(model="gpt-4o", messages=messages, tools=tools)

if response.choices[0].message.tool_calls:
    tool_call = response.choices[0].message.tool_calls[0]
    search_results = exa_search(json.loads(tool_call.function.arguments)["query"])
    messages.append(response.choices[0].message)
    messages.append({"role": "tool", "tool_call_id": tool_call.id, "content": search_results})
    final = openai.chat.completions.create(model="gpt-4o", messages=messages)
    print(final.choices[0].message.content)

Anthropic Tool Use

import anthropic
from exa_py import Exa

client = anthropic.Anthropic()
exa = Exa()

tools = [{
    "name": "exa_search",
    "description": "Search the web for current information.",
    "input_schema": {
        "type": "object",
        "properties": {"query": {"type": "string", "description": "Search query"}},
        "required": ["query"]
    }
}]

def exa_search(query: str) -> str:
    results = exa.search(query=query, type="auto", num_results=10, contents={"text": {"max_characters": 20000}})
    return "\n".join([f"{r.title}: {r.url}" for r in results.results])

messages = [{"role": "user", "content": "Latest quantum computing developments?"}]
response = client.messages.create(model="claude-sonnet-4-20250514", max_tokens=4096, tools=tools, messages=messages)

if response.stop_reason == "tool_use":
    tool_use = next(b for b in response.content if b.type == "tool_use")
    tool_result = exa_search(tool_use.input["query"])
    messages.append({"role": "assistant", "content": response.content})
    messages.append({"role": "user", "content": [{"type": "tool_result", "tool_use_id": tool_use.id, "content": tool_result}]})
    final = client.messages.create(model="claude-sonnet-4-20250514", max_tokens=4096, tools=tools, messages=messages)
    print(final.content[0].text)

Search Type Reference

Tip: type="auto" works well for most queries. It provides balanced relevance and speed.

Content Configuration

Choose ONE content type per request (not both):

⚠️ Token usage warning: Using text: true (full page text) can significantly increase token count, leading to slower and more expensive LLM calls. To mitigate:

• Add max_characters limit: "text": {"max_characters": 10000}
• Use highlights instead if you don't need contiguous text

When to use text vs highlights:

• Text - When you need untruncated, contiguous content (e.g., code snippets, full articles, documentation)
• Highlights - When you need key excerpts and don't need the full context (e.g., summaries, Q&A, general research)

Domain Filtering (Optional)

Usually not needed - Exa's neural search finds relevant results without domain restrictions.

When to use:

• Targeting specific authoritative sources
• Excluding low-quality domains from results

Example:

{
  "includeDomains": ["arxiv.org", "github.com"],
  "excludeDomains": ["pinterest.com"]
}

Note: includeDomains and excludeDomains cannot be used together.

Coding Agent

Use category: "null" to search for null content.

{
  "query": "React hooks best practices 2024",
  "category": null,
  "num_results": 10,
  "contents": {
    "text": {
      "max_characters": 20000
    }
  }
}

Tips:

• Use type: "auto" for balanced results
• Great for documentation lookup, API references, code examples

SDK Examples

Category Examples

Use category filters to search dedicated indexes. Each category returns only that content type.

Note: Categories can be restrictive. If you're not getting enough results, try searching without a category first, then add one if needed.

People Search (category: "people")

Find people by role, expertise, or what they work on

curl -X POST 'https://api.exa.ai/search' \
  -H 'x-api-key: YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
  "query": "software engineer distributed systems",
  "category": "people",
  "type": "auto",
  "num_results": 10
}'

Tips:

• Use SINGULAR form
• Describe what they work on
• No date/text filters supported

Company Search (category: "company")

Find companies by industry, criteria, or attributes

curl -X POST 'https://api.exa.ai/search' \
  -H 'x-api-key: YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
  "query": "AI startup healthcare",
  "category": "company",
  "type": "auto",
  "num_results": 10
}'

Tips:

• Use SINGULAR form
• Simple entity queries
• Returns company objects, not articles

News Search (category: "news")

News articles

curl -X POST 'https://api.exa.ai/search' \
  -H 'x-api-key: YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
  "query": "OpenAI announcements",
  "category": "news",
  "type": "auto",
  "num_results": 10,
  "contents": {
    "text": {
      "max_characters": 20000
    }
  }
}'

Tips:

• Use livecrawl: "preferred" for breaking news
• Avoid date filters unless required

Research Papers (category: "research paper")

Academic papers

curl -X POST 'https://api.exa.ai/search' \
  -H 'x-api-key: YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
  "query": "transformer architecture improvements",
  "category": "research paper",
  "type": "auto",
  "num_results": 10,
  "contents": {
    "text": {
      "max_characters": 20000
    }
  }
}'

Tips:

• Use type: "auto" for most queries
• Includes arxiv.org, paperswithcode.com, and other academic sources

Tweet Search (category: "tweet")

Twitter/X posts

curl -X POST 'https://api.exa.ai/search' \
  -H 'x-api-key: YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
  "query": "AI safety discussion",
  "category": "tweet",
  "type": "auto",
  "num_results": 10,
  "contents": {
    "text": {
      "max_characters": 20000
    }
  }
}'

Tips:

• Good for real-time discussions
• Captures public sentiment

Content Freshness (maxAgeHours)

maxAgeHours sets the maximum acceptable age (in hours) for cached content. If the cached version is older than this threshold, Exa will livecrawl the page to get fresh content.

When LiveCrawl Isn't Necessary: Cached data is sufficient for many queries, especially for historical topics or educational content. These subjects rarely change, so reliable cached results can provide accurate information quickly.

See maxAgeHours docs for more details.

Other Endpoints

Beyond /search, Exa offers these endpoints:

Example - Get contents for URLs:

POST /contents
{
  "urls": ["https://example.com/article"],
  "text": { "max_characters": 20000 }
}

Troubleshooting

Results not relevant?

1. Try type: "auto" - most balanced option
2. Refine query - use singular form, be specific
3. Check category matches your use case

Results too slow?

1. Use type: "fast"
2. Reduce num_results
3. Skip contents if you only need URLs

No results?

1. Remove filters (date, domain restrictions)
2. Simplify query
3. Try type: "auto" - has fallback mechanisms

Resources

• Docs: https://exa.ai/docs
• Dashboard: https://dashboard.exa.ai
• API Status: https://status.exa.ai