Server defines 8 tools with explicit schemas and descriptions. Naming follows verb_noun pattern (creo_get_*, creo_create_*, creo_verify_*). Descriptions are present and moderately detailed (avg ~150 chars), explaining what each tool does and its constraints. However, several parameters lack descriptions entirely, and output schemas are not documented. Tool annotations are present (readOnlyHint, destructiveHint, idempotentHint), which is good. Error handling guidance is minimal. Parameter validation is strong (enums, patterns, length limits), but LLM-facing descriptions of constraints are sometimes buried in JSON Schema rather than in parameter descriptions.

• Document output schemas for all 8 tools. Specify field names, types, and what each field represents. Example: creo_get_current_model should document {model_name: string, model_type: string, dimensions: {width: number, height: number, depth: number}}.
• Expand creo_get_project_dimensions description from 27 chars to 100-150 chars. Explain: reads dimensions from a part in the live working directory (not safe output). Clarify when to use this vs creo_get_dimensions.
• Split creo_start_resident_and_get_basic_model into two tools: creo_start_resident_bridge (idempotent, no output) and creo_get_basic_model (reads working dir, model name, type, dimensions). Agents can compose them as needed.
• Add parameter descriptions for all optional/conditional params. For creo_get_dimensions.model_file, document: 'Optional. If omitted, reads current model. If provided, reads from safe output directory.'
• Add error handling guidance to tool descriptions. Example: 'If Creo is not running, returns error code CREO_NOT_FOUND. Retry after starting Creo. If model file not found in safe output, returns FILE_NOT_FOUND with available files listed.'