XcodeMCP provides 20 well-structured tools with complete JSON schemas and descriptions. However, the server exhibits significant issues that prevent a higher score: (1) Tool naming lacks consistency, some tools use verb_noun pattern (xcode_open_project, xcode_build) but many lack clear action verbs or use redundant prefixes (xcresult_* tools repeat the resource type). (2) Descriptions are generally adequate (80-160 chars) but lack guidance on when to use which tool, prerequisites, and recovery paths. For example, xcode_test has a complex schema with interdependent parameters (selected_tests, selected_test_classes, test_plan_path) but the description does not explain the relationships or when each option applies. (3) Output schemas are completely undocumented, we can see input schemas clearly, but return types and field structures are never specified in the source code, making it impossible for LLMs to plan downstream calls or extract the right data. (4) Error handling is minimal, no error classification, no actionable recovery guidance, no suggestions for alternatives when operations fail. (5) Composition suffers: tools like xcode_build_and_run combine multiple concerns (build + run) that should be separate for agent flexibility. (6) Parameter dependencies are undocumented, xcode_test requires test_plan_path when using selected_tests or selected_test_classes, but this constraint is buried in the description, not explicitly marked as a relationship.

• Add outputSchema to every tool definition documenting the response structure (field names, types, required fields). Follow the pattern: { type: 'object', properties: { field_name: { type: '...', description: '...' } }, required: [...] }. This is critical for LLM reasoning.
• For xcode_test, explicitly document parameter relationships: 'Use EITHER selected_tests OR selected_test_classes (not both). Both require test_plan_path to be set. If neither is specified, runs all tests in the target.' Make this a required field constraint in the schema.
• Split xcode_build_and_run into two separate tools: xcode_build (returns build artifacts) and xcode_run (accepts build artifacts or scheme). This allows agents to build without running, or run without rebuilding.
• Rename xcresult_* tools to be flatter and clearer: xcresult_get_summary (instead of xcresult_summary), xcresult_get_test_console (instead of xcresult_browser_get_console), xcresult_list_test_attachments (instead of xcresult_list_attachments). This reduces cognitive load when choosing among similar tools.
• Enrich tool descriptions with discovery hints: 'Call xcode_get_schemes() first to see available schemes.' 'Call xcresult_browse() first to list test IDs before calling xcresult_get_screenshot().' This guides multi-step planning.
• Add error classification and recovery guidance to tool descriptions. E.g., 'If build fails, check the build log (returned in response.log_path). Common causes: missing dependencies (run pod install), outdated scheme (call xcode_get_schemes()), incompatible destination (call xcode_get_run_destinations())'.

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

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