wake-mcp provides 7 tools with clear naming and reasonably documented schemas. All tools follow verb_noun patterns (wake_status, wake_log, wake_list_commands, wake_get_output, wake_search, wake_dump, wake_annotate). Input schemas are explicitly visible and properly structured with JSON Schema. However, parameter descriptions are sparse or missing entirely in several tools, and output schemas are completely undocumented, a critical gap. The tool names are action-oriented and unambiguous, which is a strength. Descriptions range from adequate to good (44-261 chars), meeting baseline expectations. The primary weaknesses are: (1) no output schema documentation for any tool, violating the pattern:tool and pattern:response-shaper requirements; (2) missing or minimal parameter descriptions in several fields (e.g., session_id in wake_dump has no description despite being optional); (3) no error handling guidance, tools return results but provide no recovery paths if operations fail; (4) wake_annotate is the only write operation but lacks any confirmation/safety mechanism despite being able to mutate state.

• Document output schemas for all tools. Example for wake_list_commands: {"type":"object","properties":{"commands":{"type":"array","items":{"type":"object","properties":{"id":{"type":"integer"},"command":{"type":"string"},"exit_code":{"type":"integer"},"output_bytes":{"type":"integer"},"summary":{"type":"string"}}}},"total":{"type":"integer"}}}.
• Add descriptions to all input parameters, even optional ones. Example: wake_log count parameter should say 'Number of recent commands to retrieve (default: 10, range: 1-1000)'.
• Add error handling guidance to each tool description. Example for wake_get_output: 'If command IDs are invalid or not found, returns empty results. Call wake_list_commands to discover valid IDs.'
• Implement a confirmation step for wake_annotate via a Multi Round-Trip Request (result "input_required"). Example: first call returns {"status":"confirm","message":"Add annotation to session X?","input_required":true}, requiring user approval before writing.
• Constrain numeric parameters with explicit bounds in both schema defaults and descriptions. Change wake_log count default JSON to include minimum 1, maximum 1000. Similar for wake_list_commands.
• Document when to use each read tool in their descriptions. Example for wake_list_commands: 'Use this to discover recent commands cheaply (metadata only). Call wake_get_output afterward to fetch full output for specific commands of interest.'
• Add pagination guidance if result sets can grow large. Clarify if these tools enforce internal limits or if LLMs need to use the count parameter to paginate.