This repository contains 24 tools across 5 MCP adapters (AirPods, GitHub, Playwright, Shortcut). Critical issues: (1) All tools use generic operation-dispatch patterns (mcp_aql_read, mcp_aql_create, etc.) rather than verb_noun naming. Four tools share identical names across different domains, making them indistinguishable to LLMs. (2) Parameter schemas are severely underdeveloped: 'operation' is an enum but 'params' is a bare object with no type, properties, or required fields defined. This violates JSON Schema best practices and leaves LLMs unable to understand what parameters each operation accepts. (3) Descriptions are domain-specific but lack operation-level detail. The mcp_aql_read for AirPods lists 8 operations in the description but provides no per-operation guidance on parameters, return types, or when to use each. (4) No output schemas documented anywhere. (5) No error handling guidance. The generic dispatch pattern hides implementation details from the LLM, forcing it to reason about operation names and parameter structures without schema support.

• Rename tools to domain-specific verb_noun patterns: 'get_airpods_pose', 'list_airpods_devices', 'start_airpods_sidecar', 'list_github_repos', 'create_github_issue', 'execute_playwright_script', etc. Each tool name should clearly indicate domain and action.
• Expand parameter schemas to define operation-specific properties. For each operation enum value, document required and optional parameters with types and descriptions. Example: operation='get_pose' requires {device_id: string}, operation='list_devices' requires no params.
• Document output schemas for each operation. Add a 'Returns' section to tool descriptions specifying the structure of successful responses. Example: 'get_pose returns {x: number, y: number, z: number, timestamp: string}'.
• Add per-operation error handling guidance. Document which operations can fail, what errors they raise, and what the LLM should do. Example: 'If device not found, call list_devices first to verify available devices.'
• Consider splitting generic dispatch tools into domain-specific tools. Instead of one 'mcp_aql_read' with 8 operations, create 'get_airpods_pose', 'get_airpods_calibration', 'list_airpods_devices' as separate tools. This improves LLM discoverability and reduces parameter complexity.