Related Servers

The core fetch family is well-differentiated by the descriptions (fetch_markdown for reading, fetch_html for markup, browse_page for actions, list_elements for selector discovery), but the large set of site-specific tools (amazon_search, walmart_search, ebay_search, zillow_search, indeed_jobs, etc.) overlaps heavily with run_plugin, which can execute those same plugins. An agent must decide between a dedicated vertical tool and the generic plugin runner, and fetch_parsed vs run_plugin is also fuzzy. Descriptions mitigate but do not remove the ambiguity.

Two clear families exist: verb_noun for generic operations (fetch_markdown, fetch_html, fetch_parsed, browse_page, list_elements, list_plugins, run_plugin, search_and_read) and site_action for verticals (google_search, amazon_search, walmart_search, zillow_search, youtube_search). Minor deviations (indeed_jobs instead of indeed_search, maps_place_details, booking_hotels) are readable and predictable in context.

24 tools is on the heavy side and much of the bulk comes from site-specific wrappers that run_plugin could largely subsume, so the surface feels wider than necessary. The generic scraping primitives plus plugin discovery form a coherent small core; the remaining verticals inflate the set.

Coverage of fetching, rendering, interaction, parsing, search verticals, and major commerce/reviews/jobs/travel sites is strong and covers most scraping workflows end to end. Gaps are minor: no screenshot/PDF capture, no crawl/sitemap or batch-job tooling, and no account/usage/quota introspection.