Loading ToolBench
Coordination intelligence for Autodesk Navisworks, exposed over MCP. Analyzes clash detection data, provides coordination insights, and exposes analysis and document manipulation over the Model Context Protocol.
NavisCoord exhibits significant gaps in parameter schema documentation and output specifications. While tool names are clear action-oriented verbs (navis_health, navis_load_clash_export, navis_analyze), descriptions are present but lack the specificity needed for LLM-driven tool selection. Critical issues: (1) Input schemas are largely invisible in the provided source, only 'navis_health' and 'navis_target' show empty object inputs, but 14 tools total have no visible parameter documentation in the excerpt. (2) Output schemas are not documented anywhere in the provided code. (3) Parameter descriptions are absent from the visible schema definitions. (4) Error handling guidance is not evident. The server does show good naming consistency (all navis_* prefix, action verbs) and acknowledges risk classifications (READ_ONLY, WRITE, DESTRUCTIVE), but without visible parameter constraints, enums, format specifications, or output structure documentation, the tool definitions cannot guide LLM behavior reliably. The code comments in mcp_server.py indicate awareness of best practices ('heavy data never reaches the model', 'Writes identify their target'), but these are not reflected in the actual registered schemas visible in the source.
Ejecuta el motor de análisis de coordinación sobre el clash export en memoria.
Registra y recupera operaciones de cambio de apariencia.
Solicita la salida de la aplicación Navisworks.
Renderiza en 3D una captura de un cruce específico, con anotaciones.
Cierra o guarda el documento abierto en Navisworks.
Estado del puente y del documento abierto en Navisworks. Llama esto PRIMERO. Reporta la versión de Navisworks, qué documento está activo, cuántos modelos tiene la federación y cuántos tests y resultados de clash existen. Todo lo demás actúa sobre ese documento. Si hay varias instancias de Navisworks abiertas, responde la seleccionada con navis_target o, si no elegiste ninguna, la más reciente — y lo dice.
Input schemas are not visible in source code. No parameters documented for any of the 14 tools except navis_health and navis_target (shown as empty objects). Tools like navis_load_clash_export, navis_load_profile, navis_set_appearance should accept file paths, profiles, or configuration parameters, but these are not documented in the visible schema.
Output schemas are completely undocumented. No tool specifies what it returns, what fields are in the response, or what the LLM should expect to parse. Tools like navis_rank_by_severity, navis_hotspots, and navis_clash_image likely return complex structured data (ranked lists, spatial zones, image metadata), but the response structure is invisible to the LLM.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 42 | 2026-07-28+ | v2 |
Zonas espaciales donde se concentran los problemas de coordinación.
Lee un archivo JSON de clash export y lo analiza. El análisis permanece en memoria para llamadas posteriores y se invalidará cuando cambies de documento.
Carga un perfil de coordinación editado a mano. Valida su estructura y lo usa para posteriores análisis. El perfil está seleccionado hasta que cierres el servidor.
Información sobre el perfil cargado actualmente.
Los problemas de coordinación encontrados, clasificados por severidad.
Deshace cambios de apariencia registrados en una operación anterior.
Colorea elementos del modelo según una estrategia de paleta.
Elige qué instancia de Navisworks responde a las llamadas siguientes.
Parameter descriptions are absent. Tools accept inputs (file paths, configuration, spatial filters, appearance strategies) that require guidance for LLMs, but no parameter-level documentation is visible. E.g., navis_load_profile likely accepts a file_path parameter, what format? Maximum size? Is it relative or absolute?
Destructive operations (navis_document_close, navis_application_exit) lack dry_run or confirmation patterns. The descriptions note they are DESTRUCTIVE but provide no mechanism for the LLM to preview or confirm before execution, violating pattern:confirmation-request.
Tool descriptions lack contextual guidance. While they state WHAT the tool does, they do not explain WHEN to call it, what prerequisites exist, or how it relates to other tools. E.g., navis_analyze says 'Ejecuta el motor de análisis' but doesn't say 'call navis_load_clash_export first' or 'requires a profile to be loaded via navis_load_profile'.
Error handling is not documented. No tool describes what error conditions it can raise, whether errors are retryable, or what the LLM should do if a call fails. E.g., navis_load_clash_export might fail with 'invalid JSON format' or 'file not found', the LLM has no guidance.
Descriptions are in Spanish, which is appropriate for Spanish-speaking users, but inconsistent localization reduces clarity for multi-language agent ecosystems. Some descriptions are under 50 characters (navis_target: 'Elige qué instancia...'), which may be too terse for LLM-driven selection in a complex tool palette.