Single tool 'send-email' has a well-structured schema with proper email format validation and comprehensive parameters. Descriptions are present and moderately detailed (avg ~150 chars per param). However, several parameters include prescriptive LLM instructions ('You MUST ask the user...') which belong in system prompts, not tool descriptions. Conditional parameter requirements (from, replyTo) are documented but could be clearer. Output schema is implicit (returns text content) but not formally documented. Error handling is basic (throws on missing from/replyTo, API errors) without recovery guidance.

• Move LLM instructions ('You MUST ask the user...') from parameter descriptions into the client's system prompt. Keep tool descriptions focused on WHAT the parameter does, not HOW the LLM should use it.
• Explicitly document conditional requirements: 'from: Required if SENDER_EMAIL_ADDRESS env var is not set. Format: valid email address.' Make the condition clear in the description text.
• Add formal output schema documentation to the tool definition. Document that the response is {content: [{type: 'text', text: string}]} so LLMs know the structure before calling.
• Enhance error messages with recovery paths. E.g., 'Missing from address. Set SENDER_EMAIL_ADDRESS environment variable or pass from parameter with a valid email.' This guides the LLM to self-correct.
• Document timeout and retry behavior. Add to tool description: 'This tool calls the Resend API (timeout: 30s). Transient failures (429, 5xx) are retryable; permanent failures (invalid email, auth) are not.'