FadCat provides 8 tools for Android debugging via ADB/logcat. Tool definitions are present with names, descriptions, and input schemas visible in src/mcp/server.py. However, several critical gaps prevent a higher score: (1) descriptions are minimal (avg ~50-80 chars, below the 194-char production baseline), lacking WHEN-to-use and recovery guidance; (2) parameter descriptions exist but are terse and lack format/constraint details (e.g., 'Device serial number' without serial format); (3) output schemas are not documented in the tool definitions, the code references Pydantic models (LogLine, ProcessInfo, StackTraceInfo, etc.) but these are not exposed as JSON Schema in tool registration; (4) no explicit error handling guidance, tools may fail but responses don't tell the LLM what to do next; (5) all tools are READ_ONLY and safe, but this is not annotated in the schema (tool annotations are claimed in FEATURES but not visible in tool definitions). The server uses fastmcp, which is modern, but tool definitions appear minimal in the provided code snippet.

• Document output schemas as JSON Schema in each tool definition. Map Pydantic models to JSON objects with explicit field types and descriptions. E.g., 'Returns: {"devices": [{"serial": string, "model": string, "status": string}]}'.
• Expand descriptions to 100-200 chars. Add WHEN-to-use guidance. E.g., 'Get list of connected Android devices. Call this at startup or when you need to switch devices. Returns device serial numbers and model names.'
• Convert parameter descriptions to include format and constraints. E.g., 'device: Device serial number (format: emulator-NNNN or alphanumeric ID from get_devices())'.
• Add formal enum constraints for 'level' parameter in get_logcat_stream and filter_by_level: {"type": "string", "enum": ["V", "D", "I", "W", "E", "F"]}.
• Add min/max constraints for numeric parameters: 'lines' (1 - 10000), 'duration' (1 - 300 seconds).
• Consolidate filter_by_level into get_logcat_stream or rename to clarify its state model. If stateless, document how it differs. If stateful, redesign to avoid hidden session dependencies.
• Add error handling guidance in tool descriptions or docstrings. E.g., 'If device not found, call get_devices() to refresh the list.'
• Add readOnlyHint to tool schemas: {"readOnlyHint": true} to all 8 tools. This tells clients these are safe to retry.
• Validate device serial and package name formats before passing to ADB. Return actionable errors: 'Device "unknown" not found. Available: emulator-5554, device-serial-1'.
• Add pagination support to search_logs and get_app_processes if results can be large (LLM context risk). Return 'has_more' flag and 'next_cursor' for continuation.

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

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