Tools have basic schemas and descriptions, but quality is inconsistent. Most descriptions are terse (10-50 chars), below the 50-200 char baseline for LLM-optimized docs. Parameter descriptions are sparse or missing entirely. web_fetch and web_docs have moderately detailed schemas; others lack depth. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all being read-only. Error handling and recovery guidance absent. Output schemas undocumented. Tool names follow verb_noun convention (web_search, web_fetch) but lack specificity, 'web_fetch' doesn't clarify what 'fetch' returns or when to use it vs web_links.

• Expand tool descriptions to 50-200 chars. Example: 'web_search' → 'Search the web using multiple queries (1-4). Returns ranked results with titles, snippets, and URLs. Use for broad research; use web_fetch for full page content.'
• Document all parameter enums with explanations. For web_fetch 'mode': 'auto (default): smart selection | full: entire page content | tree: DOM structure only'. For 'render': 'http: fast, text-only | browser: JavaScript-enabled, slower'.
• Add output schema documentation to each tool. Example: 'web_search returns {results: [{title, snippet, url, rank}], total_count, next_cursor}'.
• Add readOnlyHint: true to all tool definitions (via tool annotations in MCP protocol). This signals to agents that these tools are safe to call without confirmation.
• Implement structured error responses with recovery hints. Example: 'web_search failed: rate limited. Retry in 60s or try web_docs for cached documentation.' Use pattern:recovery-guide.