MXCP provides 11 tools across Confluence, Jira, and internal utilities. While all tools have descriptions and most have structured input schemas, there are significant quality gaps: (1) descriptions lack LLM-optimization depth and context on when to use each tool vs alternatives; (2) several parameter descriptions are minimal or missing constraint details; (3) no visible output schemas or return type documentation; (4) error handling and recovery guidance is not evident in tool definitions; (5) no tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being READ_ONLY; (6) composition issues, e.g., search_pages and cql_query both search but differ in interface, risking LLM confusion. The server demonstrates basic structure but falls short of production-grade tool definition quality expected in pattern:tool and pattern:tool-description.

• Document output schemas for all tools. For list_spaces, specify return type (array of {key, name}). For get_page, document whether content is HTML, markdown, or raw storage format.
• Add tool annotations: all tools should declare readOnlyHint=true. E.g., { name: 'get_page', inputSchema: {...}, readOnlyHint: true }
• Expand search_pages and cql_query descriptions to clarify: 'search_pages: keyword-based search of page titles/bodies (simpler, faster). cql_query: advanced CQL queries for filtering by metadata (labels, created date, author). Use search_pages for simple keyword searches; cql_query for complex filters.'
• Add min/max bounds to numeric parameters: limit (min=1, max=100, default=10) and max_results (min=1, max=100, default=50). Document pagination: 'To retrieve more results, use offset parameter or cursor from previous response.'
• Enhance parameter descriptions with examples and constraints. E.g., cql_query: 'The CQL query string (e.g., "type=page AND label=draft"). See Confluence CQL syntax docs.'
• Add per-tool error handling. E.g., get_page: 'Returns 404 if page_id is invalid. Call search_pages first if you only have a keyword. Returns 403 if user lacks read permission, recommend suggesting related pages instead of failing.'
• For list_spaces, add optional pagination: limit (default=50, max=100), offset. Document: 'If more than limit spaces exist, call again with offset=previous_limit to fetch the next batch.' Return total_count for client-side pagination.

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

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