| name | web-research |
| description | Use this when the task needs public-web reading, webpage inspection, or link extraction, especially after getting candidate URLs from an MCP search tool. |
| metadata | {"short-description":"Public web reading and MCP-assisted search workflow"} |
Use the built-in web tools for public websites:
fetch_url
read_webpage
extract_page_links
Preferred workflow:
- If an MCP search tool is available, use it first to find relevant URLs.
- Use
read_webpage for human-readable page text and metadata.
- Use
fetch_url when the raw textual payload matters, such as JSON, XML, or HTML source.
- Use
extract_page_links to gather follow-up URLs from a page you already trust.
Limits and guardrails:
- These tools are read-only and only support public
http and https targets.
- Localhost and private-network targets are blocked unless the runtime explicitly allows them.
- There is no built-in search tool in core v1. If no MCP search provider is configured, say search is unavailable and ask for a direct URL instead.
- Do not assume JavaScript rendering, browser tabs, cookies, login state, screenshots, or file downloads.