v4.0.0: slim tool defs and responses for ~30-55% smaller MCP payloads

[ttpears]

Summary

Token-efficiency overhaul. Verified against a live BookStack via scripts/bench-context.mjs:

Endpoint	Before	After	Saved
tools/list	13,636	9,290	32%
search_content (5)	17,735	10,985	38%
get_books (5)	3,636	1,570	57%
get_pages (5)	4,826	2,169	55%
get_recent_changes (5)	19,302	9,519	51% (also no N+1)
get_shelves (5)	4,700	2,090	56%
get_page (1)	6,333	5,779	9% (content-bound)
get_book (1)	12,974	8,953	31%

Response shape changes (breaking)

• Drop redundant fields: direct_link, *_friendly date strings, content_info, contextual_info, change_summary, pagination_hint, location, shelf summary / tags_summary / book_count.
• Fix buggy attachment page_url; keep download_url.
• Compact JSON (no pretty-print whitespace).
• get_recent_changes returns search-response data directly — no per-result book/page/chapter fetches.

Tool-def changes

• Remove get_capabilities (redundant with tools/list).
• Drop title field (clients key on name).
• Trim verbose descriptions and 52 boilerplate param descriptions.
• Dedupe advanced-search caveat onto search_content only.

Other

• Add scripts/bench-context.mjs JSON-RPC harness for re-measuring payload sizes against a live BookStack.
• README: 4.0.0 breaking-changes callout, feature bullet for the size win, remove get_capabilities row.
• Version bumped in package.json and gemini-extension.json.

Test plan

• npm run type-check passes
• npm run build passes
• Live benchmark run against bookstack.teamgleim.com (read-only, sizes above)
• Spot-checked response shapes for search_content, search_pages (advanced syntax {type:page}), get_books, get_pages, get_shelves, get_page (content pagination intact), get_book (chapter/page hierarchy intact), get_recent_changes
• CI: build job green on PR
• After merge: tag job creates v4.0.0, release workflow publishes to npm with provenance