web-search
Search the web using Exa AI for real-time information retrieval. Use when users ask to search online, find current information, look up recent events, or need data beyond knowledge cutoff. Triggers on "search the web", "find online", "look up", "what's the latest on", "search for".
When & Why to Use This Skill
This Claude skill enables real-time web searching and information retrieval using Exa AI's specialized MCP endpoint. It allows Claude to bypass its training data cutoff by accessing the live web, providing structured JSON results and source URLs optimized for LLM consumption. It is essential for tasks requiring up-to-the-minute accuracy, such as news tracking, technical documentation lookups, and deep academic or market research.
Use Cases
- Real-time News and Events: Retrieving the latest information on breaking news, sports scores, or current global events that occurred after the model's knowledge cutoff.
- Technical Documentation Lookup: Finding the most recent API references, library updates, and software documentation to ensure coding tasks use the latest standards.
- Fact-Checking and Verification: Verifying claims or data points by searching for primary sources and authoritative references across the live web.
- Deep Market Research: Conducting comprehensive searches on industry trends, competitor updates, or niche topics using the 'deep' search mode for detailed analysis.
- Academic and Professional Sourcing: Locating specific papers, articles, or reports to provide high-quality citations and evidence-based answers.
| name | web-search |
|---|---|
| description | Search the web using Exa AI for real-time information retrieval. Use when users ask to search online, find current information, look up recent events, or need data beyond knowledge cutoff. Triggers on "search the web", "find online", "look up", "what's the latest on", "search for". |
Web Search
Overview
Search the web using Exa AI's MCP endpoint for real-time information retrieval. Returns raw JSON results with source URLs optimized for LLM consumption.
When to Use
- User asks to search the web or find online information
- Need current/recent information beyond knowledge cutoff
- Looking up documentation, news, or real-time data
- Researching topics that require up-to-date sources
Workflow
- Identify the search query from user request
- Determine appropriate search parameters:
--type auto(default) for balanced results--type fastfor quick lookups--type deepfor comprehensive research
- Run the search script
- Present results with source URLs to user
Usage
# Basic search
uv run --script scripts/web_search.py --query "your search query"
# With options
uv run --script scripts/web_search.py \
--query "search query" \
--num-results 8 \
--type auto \
--livecrawl fallback \
--context-max-chars 10000 \
--timeout 25
Parameters
| Parameter | Default | Description |
|---|---|---|
--query |
(required) | Search query string |
--num-results |
8 | Number of results to return |
--type |
auto | Search type: auto, fast, deep |
--livecrawl |
fallback | Live crawl mode: fallback, preferred |
--context-max-chars |
10000 | Max characters for context |
--timeout |
25 | Request timeout in seconds |
Output Contract
| Scenario | stdout | stderr | exit code |
|---|---|---|---|
| Success | Raw JSON from Exa | (empty) | 0 |
| No results | {"results": []} |
Warning message | 0 |
| Error | (empty) | Error message | 1 |
Success output contains:
- Page titles and URLs
- Content snippets optimized for LLM context
- Source attribution
Prerequisites
- Uses Exa AI's free MCP endpoint (no API key required)
- Requires
uvfor running PEP 723 scripts
Examples
Quick lookup
uv run --script scripts/web_search.py \
--query "Python 3.12 new features" \
--type fast \
--num-results 3
Deep research
uv run --script scripts/web_search.py \
--query "LLM agent architectures 2024" \
--type deep \
--num-results 10 \
--livecrawl preferred