test-genie-mcp has 21 tools with visible schemas and descriptions. Naming follows verb_noun convention consistently (analyze_, generate_, create_, run_, detect_, suggest_, apply_, etc.). However, several critical issues reduce overall quality: (1) Many descriptions lack specificity about prerequisites, return structures, or error conditions. (2) Parameter descriptions are present but often generic (e.g., 'Path to the project root' appears 20+ times without explaining why projectPath is required for each distinct tool). (3) No documented output schemas, tools return JSON but LLMs cannot plan what fields to expect. (4) Several tools combine multiple concerns (run_full_automation does analyze→scenarios→test→detect→fix→report in one call; run_iterative_fix_loop repeats test→detect→fix cycle). (5) STDIO transport is a hard ceiling at 50 per spec rules, but definition quality alone is mediocre-to-fair. Positive aspects: all tools have names, descriptions, and input schemas with types; enums are used appropriately for constrained inputs (platform, testType, coverage, etc.); parameter descriptions exist for all inputs. The server is functional but would benefit from output schema documentation, tighter tool composition, and richer error handling.

• Document output schemas for every tool. Include example JSON responses and field descriptions. E.g., analyze_app_structure should document that it returns {screens: [...], components: [...], apis: [...], state: {...}, detectedPlatform: string}. This enables LLMs to chain tools based on known field names.
• Split run_full_automation and run_iterative_fix_loop into single-purpose tools. Instead of one mega-tool, offer: orchestrate_test_workflow (accepts config) that internally chains smaller tools. Or expose smaller pieces (run_tests, batch_detect_issues, batch_apply_fixes) so agents can compose.
• Enrich parameter descriptions with context specific to each tool. Instead of 'Path to the project root' for all 21 tools, explain: 'analyze_app_structure uses this path to scan source files and config. Must contain package.json (Node/web) or AndroidManifest.xml (Android) or equivalent.'
• Add error handling guidance to every tool description. For irreversible operations (apply_fix, run_full_automation), document: 'If fix application fails, call rollback_fix(fixId) to undo changes.' For detection tools, document: 'Returns empty array if no issues found, not an error.'
• Document whether each tool depends on prior tools. E.g., does generate_scenarios require analyze_app_structure to run first? Does run_scenario_test require create_test_plan? Clarify in descriptions so agents know the correct sequence.
• Add structured tool annotations (readOnlyHint, destructiveHint, idempotentHint) to distinguish safe reads from risky writes. Replace markdown [mode: real/hybrid/simulated] tags with formal capability extensions negotiated via _meta.