| name | web-search-provider |
| description | Search the web and extract content from URLs |
| min_capability | tier-2 |
| requires | {"connectors":["web"]} |
Web Search
You can now search the web and extract content from URLs.
How to handle search requests
When a user asks to:
- "Search for the latest news about AI"
- "What's the weather in Buenos Aires?"
- "Find information about Python 3.13"
- "Look up the best restaurants in Madrid"
Do the following:
- Identify the search query from the user's message
- Use web__search with the query
- Summarize the results in a clear, concise way
- If the user wants more detail, use web__extract on a specific URL
Best practices
- Keep summaries concise — 2-3 paragraphs max
- Cite sources when providing factual information
- If search returns no results, suggest refining the query
- For time-sensitive info (weather, news), note that results may not be real-time
Limitations
Web search depends on the web connector handler being configured.
For full functionality, consider connecting an MCP server like
@anthropic/mcp-server-fetch or using a search API (Tavily, Brave).