Single tool 'search' has a verb-based name (good), but critical gaps undermine quality. Description is only 6 characters ("搜索关键字" = "Search keyword"), far below the 10-1024 character baseline and provides no context for LLM selection. Input schema exists but lacks parameter description, the 'query' field has no explanation of expected format, length, or constraints. Output is unstructured plain text (concatenated search result snippets), with no documented schema, pagination, or field structure. No error handling guidance, failures return bare False instead of actionable recovery messages. Tool does not follow the response-shaper or paginated-result patterns.

• Expand tool description to 50-150 characters: 'Search the SearXNG instance for web results. Returns up to 20 snippets with titles and URLs. Use for general web queries, news, or academic content.'
• Add parameter description for 'query': 'Search query string (1-500 characters). Supports boolean operators (AND, OR, NOT) and quoted phrases. Language auto-detected from query.'
• Return structured JSON with fields: {results: [{title, url, snippet, source}], total_results, query, search_time_ms}. Implement pagination via limit/offset parameters.
• Replace bare False errors with actionable messages: 'SearXNG instance unreachable (https://searxng.abc.com). Check SEARXNG_URL env var and network connectivity.' or 'Invalid query: contains forbidden characters. Use alphanumeric, spaces, and quotes only.'
• Validate query length (1-500 chars) and sanitize special characters before URL interpolation. Use urllib.parse.quote() instead of raw f-string formatting.

Inferred effective spec: 2026-07-28+.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained