cogni-mcp-server has 9 tools with basic descriptions and minimal input schemas. Tool names follow verb_noun convention (storePose, listPoses, moveTo, etc.), which is good. However, descriptions are terse (10-60 chars), parameter descriptions are minimal or absent, and output schemas are not documented. No error handling guidance, no parameter constraints (enums, ranges), and no indication of idempotency or retry safety. The server targets robotics (arm control, pose management) but lacks the structured output and error recovery patterns needed for reliable agent integration.

• Expand tool descriptions to 100-200 chars. Example: 'storePose: Save the current robot arm configuration under a semantic label (e.g., "home", "pick_position"). Returns all joint angles in degrees. Use before moveTo to create waypoints.'
• Document output schemas for all tools. E.g., listPoses returns {poses: [{label: string, joint_angles: number[]}], total: number}. This enables LLMs to chain tools correctly.
• Add parameter constraints: label must be 1-50 alphanumeric chars; labels array must have 1-10 items. Use enums if only certain labels are valid.
• Implement confirmation pattern for destructive tools: add a 'confirm' boolean parameter or return a confirmation_required error with a confirm_delete_pose tool.
• Document idempotency and retry safety for each tool. E.g., 'moveTo is idempotent: calling it twice with the same labels produces the same final pose.' This guides agent retry logic.

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

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