Server has well-structured tool definitions with complete input schemas and descriptions for all 8 tools. Naming follows verb_noun convention (download_*, upload_*, compare, ping, get_page_content). Descriptions are detailed and explain state-modifying behavior. However, parameter descriptions lack constraint details (ranges, formats, enums), output schemas are not formally documented, and error handling guidance is minimal. Tool composition is sound, each tool has a single responsibility. The McpToolResult envelope pattern is good, but LLMs cannot see the output structure from the tool definition alone.

• Document output schemas for all tools. Add to tool descriptions: 'Returns: {success: bool, summary: string, report?: SyncReport, logs: string[]}. SyncReport contains {filesCreated, filesUpdated, filesDeleted, conflicts, errors}.' This lets LLMs understand what data is available.
• Add constraint details to parameter descriptions. E.g., 'outputDir: Local directory path (relative paths resolve against --root-dir, absolute paths must be inside sandbox). Max 260 chars.' Include format hints for spaceKey, pageId, pageTitle.
• Enhance error messages with recovery guidance. Catch specific exceptions (e.g., PageNotFound, AuthenticationFailed, PathOutsideSandbox) and return: 'error: "Page not found. Try confluence_compare to see available pages."' instead of generic messages.
• Use JSON Schema oneOf to enforce mutually exclusive parameters. Replace free-form descriptions with schema constraints: {oneOf: [{required: [pageId]}, {required: [pageTitle]}]}. This prevents invalid combinations at parse time.
• Add tool annotations (readOnlyHint, destructiveHint) to schema. Mark confluence_download_* and confluence_compare as readOnlyHint=true; mark confluence_upload_update and confluence_upload_create as destructiveHint=true. This helps agents reason about side effects.

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

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