Strong tool definitions with clear naming (verb_noun pattern), comprehensive descriptions (avg 250+ chars), and complete input schemas. All 8 tools have types and descriptions. However, output schemas are not explicitly documented in the source, and error handling guidance is minimal. No tool annotations (readOnlyHint, destructiveHint) despite all tools being read-only. Parameter descriptions are detailed but some lack explicit constraints (e.g., max_results range not stated in schema). Composition is clean, each tool has one responsibility.

• Add explicit output schema documentation for each tool. E.g., search_web returns list[{title: str, url: str, snippet: str}]. Document all fields the LLM will receive.
• Add tool annotations to FastMCP tool registration: @mcp.tool(annotations={'readOnlyHint': True}) for all 8 tools to signal safety to agents.
• Specify parameter constraints in schema: search_web max_results should have minimum=1, maximum=30 in the JSON Schema; transcribe_audio model should have enum=["tiny.en", "base.en", ...] in schema, not just description.
• Add error handling guidance to descriptions. E.g., fetch_page: 'If Cloudflare blocks the request, try render_js=false. If the page requires login, the tool will return an error, consider providing credentials via a separate auth tool.'
• Document result limits and pagination. E.g., search_web: 'Returns up to max_results items. For large result sets, implement pagination by adjusting the query (e.g., add page number to search terms).' download_media: 'If file exceeds max_mb, returns error with file size, increase max_mb or request a lower quality format.'