Single tool with minimal but present schema and description. Tool name 'fetch_and_convert' uses verb_noun pattern correctly. Description is 97 chars (within 10-1024 range) but lacks WHEN/WHY context for LLM selection. Input schema declares 'url' as string with description, but lacks format constraints (e.g., must be valid HTTP/HTTPS URL). No output schema documented, LLM cannot predict return fields. No error handling guidance. Tool is read-only (safe), but lacks idempotent/readonly annotations. Composition is single-purpose (good), but parameter validation and error recovery are absent.

• Document output schema in tool description or via FastMCP schema annotation: 'Returns {url: string, markdown: string, content_hash: string, bytes_sha256: string, status_code: int, fetched_at: ISO8601}'. This enables LLM to chain results to downstream tools.
• Add format constraint to 'url' parameter: 'A valid HTTP or HTTPS URL (e.g., https://example.com). Must be publicly accessible; private/internal URLs are blocked for security.'
• Expand tool description to 150-200 chars: 'Fetch a URL and convert its HTML to clean Markdown with byte-level provenance (content hash, SHA256, status code, timestamp). Use this to retrieve and process web content for RAG or knowledge bases. Supports public URLs only; private networks are blocked.'
• Add error recovery hints: 'If the URL is blocked (403/SSRF), try a different domain. If content is too short (<150 chars), the tool returns low_content=true, consider retrying or using a different URL.'
• Declare tool as read-only and idempotent via FastMCP annotations or tool metadata so LLM knows it is safe to retry and does not modify state.