Multi-tool MCP server for integrating 3DExperience/CATIA, Cameo, Enovia, Excel, GitLab, Nastran, PDF, and Simulink models with parameter extraction, artifact management, and simulation capabilities
Scoring was not performed
Return type schemas not documented. Tools return strings, objects, or lists but the response structure is not explicitly documented. This prevents LLMs from planning downstream tool calls or extracting the right fields for chaining.
Destructive operations lack dry-run, confirmation, or idempotency guidance. Tools like update_3dx_model_parameters, replace_engineering_instance, update_element_tags, execute_pipeline, and execute_nastran_simulation modify external systems (CATIA, Enovia, PLM databases) without confirmation or rollback patterns.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 13 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 49 | - | v1 |
Async operations are not properly exposed. Tools like extract_3dx_model_artifacts, extract_3dx_model_parameters, execute_pipeline, extract_nastran_input, execute_nastran_simulation internally call submit_job() and wait_for_job(). The async flow, timing behavior, and job state transitions are hidden from the description.
Parameter descriptions lack actionable constraints and examples. Most parameters (e.g., model_id, item_id, config_file) have trivial descriptions like 'The ID of the model' without format hints, valid range, or how to obtain them. No enum constraints for constrained parameters.
Error handling is implicit. No tool descriptions mention retryability, expected failure modes, or recovery paths. When tools fail (e.g., artifact not found, job fails, invalid model_id), the LLM has no guidance on whether to retry, call a discovery tool, or ask the user.
No pagination or result limits documented. Tools like find_engineering_items, find_documents, find_issues accept max_items but don't document default behavior, return count, or next_cursor patterns for large result sets.
Tool descriptions are inconsistently detailed. Some tools (e.g., get_nastran_results) provide bullet-point details about what fields are returned; others (e.g., replace_engineering_instance, extract_pdf_artifacts) are generic one-liners. This inconsistency makes discovery harder and leaves ambiguity about what each tool actually does.
Multi-step workflows are not decomposed. For example, to update a 3DX model parameter, the agent must call extract_3dx_model_parameters (async) → wait for job → call update_3dx_model_parameters. The tool descriptions don't hint at these dependencies or sequencing requirements.