Skip to main content
Search the web and optionally extract structured data. 
<web-search>latest bun release notes</web-search>
Returns web text plus sources, and may include structured data output when schema-driven extraction is used. Web search is available to the team lead, explorers, planners, builders, and debuggers.

​
Provider routing

Web search is routed through your team lead’s provider using their native search capabilities:
ProviderSearch support
AnthropicNative
OpenAINative
Google (Gemini)Native
Bedrock / Vertex (Anthropic models)Routed through Anthropic
If your primary provider doesn’t support native web search (OpenRouter, Vercel, Cerebras, GitHub Copilot, Local), you can route search through a supported provider by setting: 
export MAGNITUDE_SEARCH_PROVIDER=anthropic  # or "openai" or "gemini"
This requires valid credentials for the search provider (e.g. ANTHROPIC_API_KEY if routing through Anthropic).

​
Structured output

Web search can also be used with schema-driven extraction so results are parsed into a specific data shape. This is used internally by research agents to gather structured information from the web.

​
Web Fetch

Fetch the content of a specific URL. When the agent finds a link from search results and wants to read the full page, it uses this tool. 
<web-fetch>https://docs.example.com/api-reference</web-fetch>
Uses curl with a native fetch() fallback. Strips <script> and <style> from HTML. 5 MB max, 30s timeout, 5 redirect limit. Sites with aggressive bot detection (Cloudflare JS challenges) may still block requests.