feat!: remove the discover tool and Discover API integration

[nirsha-brd]

Removes the discover tool and its integration with the Bright Data Discover API (POST/GET https://api.brightdata.com/discover).

This is the inverse of #130, which added the tool.

Why this is not a leaf removal

discover was listed in base_tools (tool_groups.js), which is spread into every tool group, and separately in pro_mode_tools (server.js) — which despite its name is the set served when neither GROUPS nor TOOLS is configured, i.e. the default. It was therefore exposed in every out-of-the-box configuration, so removing it changes the default tool set clients see. Hence the major version bump.

Changes

File	Change
server.js	Deleted the discover tool definition (119 lines); dropped 'discover' from pro_mode_tools
tool_groups.js	Dropped 'discover' from base_tools
manifest.json	Removed the discover tools entry
assets/Tools.md	Removed the table row
README.md	Removed the tool row, the routing bullets, the comparison row, three use-case table rows; adjusted tool counts and the #tools-reference-* anchor; renumbered the Tool Selection list
CHANGELOG.md	### Removed entry under 3.0.0
package.json / package-lock.json	2.11.1 → 3.0.0

No changes to test/, mcp-evals/, examples/, or server.json — none reference discover.

Version: 3.0.0, following the precedent of ## [2.0.0], which was cut for "Removed duplicate web_data_ tools" — the same class of change.

Verification

Test suite unchanged from baseline: 16/16 pass before and after.

tools/list over stdio, verified via both the MCP SDK client and a raw JSON-RPC handshake:

Config	Tools before	Tools after	discover
PRO_MODE=true	74	73	absent
default (no GROUPS/TOOLS)	5	4	absent
GROUPS=research	5	4	absent
GROUPS=ecommerce,browser	28	27	absent
TOOLS=extract,scrape_as_html	2	2	absent

A set-difference of the tool names before and after yields exactly REMOVED: discover, ADDED: none — no collateral. Every group still contains the remaining base tools. Calling discover on a stale client returns a clean -32601 Unknown tool: discover rather than crashing the server. Server stderr is byte-identical across configs (only the expected dummy-token zone warning). node --check passes on all changed JS; all JSON files re-validated; every README internal anchor still resolves; all markdown tables keep header/separator/column consistency.

Remaining discover matches in the tree were each reviewed and intentionally left: the English verb in manifest.json's long_description and in prompts.js, the README's "Market discovery" / "discovery queries" phrasing, the historical CHANGELOG entry for #142, and discover_new / discover_by in server.js — the latter are Web Scraper API dataset-trigger parameters for an unrelated product. Searches for relevance_score, filter_keywords, remove_duplicates, include_content and api.brightdata.com/discover return zero hits repo-wide.

Migration

Use search_engine / search_engine_batch to find sources, and scrape_as_markdown / scrape_batch to read them.

Pre-existing issues I deliberately did not fix

Flagging rather than fixing, to keep this diff scoped to the removal:

1. README tool count is wrong independently of this change. It advertised "69 tools" while the server actually registers 74. I applied only the −1 delta (69 → 68); the real count is now 73. The README's internal arithmetic is self-consistent either way.
2. manifest.json version drift. It sits at 2.10.0 and was already stale against 2.11.1 on main. Unlike server.json, which .github/workflows/publish-mcp.yml syncs from package.json at publish time, manifest.json has no sync step. This PR edits manifest.json but leaves its version alone.
3. README.md:381 claims search_engine_batch and scrape_batch are always-enabled base tools, but base_tools never contained them — with GROUPS set they load only via advanced_scraping.

Out of scope

brightdata/skills still ships a discover-api skill, and its live-research and rag-pipeline skills are built on Discover. Those need a follow-up PR.