Starbridge exposes 16 tools with minimal quality attributes. All tool definitions are inferred from test files rather than directly visible in source code (tests/starbridge/mcp/starbridge_mcp_server_test.py), which caps per-tool assessment at 50. Most critically: no input schemas are visible for 10 of 16 tools (starbridge_claude_health, starbridge_claude_info, starbridge_claude_restart, starbridge_confluence_health, starbridge_confluence_info, starbridge_confluence_space_list, starbridge_hello_health, starbridge_hello_info, starbridge_hello_pdf, starbridge_hello_bridge), resulting in schema score of 0 for those tools. Six tools have minimal schema documentation (only parameter names and descriptions visible, no type constraints, enums, or output schemas). Descriptions are present but generic (8-120 characters), falling below the 194-char baseline. No evidence of error handling guidance, parameter validation rules, or idempotent-operation patterns. No output schema documentation visible. Tool naming follows verb_noun convention (get_, create_, delete_, list_, update_) but lacks dependency hints and multi-step guidance in descriptions.

• Move tool definitions from test files into explicit, discoverable server registration. Provide full JSON Schema for each tool input (types, constraints, enums, min/max) and documented output schemas.
• Expand tool descriptions from 50-120 chars to 150-250 chars, following WHAT/WHEN/WHY pattern. Example: 'Get health status of all services and their dependencies. Call this to verify system readiness before performing critical operations. Returns status (healthy|degraded|unhealthy) and per-service details including response time and last check timestamp.'
• Add parameter constraints: mark 'space' as an enum or pattern (e.g. /^[A-Z]{2,}$/); set max length on title (e.g. max 255) and body (e.g. max 1000000). Include examples of valid values in descriptions, not parameter names.
• Document output schemas for all tools. For list_pages, specify: {pages: [{page_id: string, title: string, space_key: string, created_at: ISO8601}], total_count: number, next_cursor?: string}. This enables LLM chaining.
• Add pagination to list tools: accept limit (1-100, default 20) and offset/cursor parameters. Return total_count and next_cursor in response to support large result sets without context explosion.
• Implement error handling: destructive operations (delete, restart) should return clear error messages with recovery hints. Example: 'Page not found. Call starbridge_confluence_page_list to find valid page IDs in this space.' Categorize errors (retryable, user_fixable, fatal).
• Add confirmation-request pattern for destructive operations: return {success: false, requires_confirmation: true, confirmation_message: 'This will permanently delete page XYZ. Confirm by calling tool again with confirm=true'} on first call.