Close.com MCP server has 20 tools with consistent naming (verb_noun pattern) and basic descriptions. All tools have input schemas with type definitions and parameter descriptions. However, descriptions are often generic (10-50 chars), output schemas are not documented, and error handling lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Parameter descriptions could be more actionable (e.g., date format examples, enum constraints). Composition is sound, tools are single-responsibility and chainable via IDs.

• Add output schema documentation to every tool. For search_leads, document: 'Returns array of lead objects with fields: id (string), name (string), company (string), email (string), phone (string), status (string).' This enables LLMs to extract and chain data.
• Add tool annotations to all state-modifying tools. Wrap create_task, update_task, delete_task, bulk_update_tasks with destructiveHint: true in the server.tool() registration. This signals to LLMs that these calls have irreversible consequences.
• Enhance error messages with recovery guidance. Instead of 'Error searching leads: <error>', return 'Lead search failed: <reason>. Try: (1) verify query syntax, (2) use search_contacts if searching by person name, (3) call test_connection to verify API access.'
• Expand tool descriptions to 80-150 characters with context. E.g., 'Search for leads by company name, contact email, or custom field. Returns up to 20 results with ID, name, company, and status. Use get_lead_details for full contact info and activity history.'
• Specify date format in parameter descriptions. Change 'Filter by date created after (ISO format)' to 'Filter by date created after (ISO 8601, e.g., 2024-01-15T10:30:00Z). Timezone defaults to account setting if omitted.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry