Scored across 6 tools
Top-level tools are cleanly separated by intent (search finds URLs, extract reads content, media handles files, config manages the server, help documents everything), and descriptions explicitly cross-reference each other. Minor overlap remains: extract's `agent` action bundles search+extract, and config__open_relay overlaps the config tool's setup domain.
Names and all internal actions are consistently lowercase snake_case (docs_query, cache_clear, extract_structured, setup_sync). The one deviation is `config__open_relay`, whose double-underscore namespace prefix breaks the single-word pattern of every other top-level tool.
Six top-level tools sit squarely in the ideal 3-15 range and are well-scoped via an action-based dispatch pattern. However this count is somewhat deceptive, since extract (9 actions), search (7), and config (11) each pack a large hidden surface, making the set heavier than six names suggest.
The surface covers the full web-toolkit lifecycle: discovery (search/research/docs/similar), retrieval and parsing (extract/crawl/map/convert/structured), interaction (interact), change tracking (diff), media (list/download), and server management (config/help). Gaps are minor, e.g. no dedicated result-history or cache-inspection tool beyond config's status.