Model Context Protocol server for Microsoft Dynamics 365 Business Central via WebUI protocol
Three tools with partial schema coverage and inconsistent description quality. get_page_metadata has a solid description (180+ chars) and clear schema with required fields. execute_action has good description but uses oneOf constraint that may confuse LLMs; schema is present but complex. create_record_by_field_name has excellent description (250+ chars) but schema lacks explicit type constraints on the 'fields' object values. All tools lack error handling guidance, recovery hints, and actionable error messages. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels documented in metadata. Parameter descriptions are present but generic; none explain format constraints, ranges, or dependencies. Output schemas are not documented, LLMs cannot predict response structure.
Creates a new Business Central record using field names/captions (e.g., "Name", "Address", "Email"). Automatically resolves field names to control paths using FormState metadata. Inputs: pageId (number|string), fields (object with field names as keys). Supports scoped fields like "General > Name" or "Address/City". Use [SourceExpr] syntax to target specific AL fields. Returns: {success, formId, setFields, failedFields}. More robust than create_record with automatic field resolution and proper oldValue handling.
Executes an action (button click) on a Business Central page. Requires pageContextId from get_page_metadata. Supports: actionId or captionPath. For subpages use target:{partId}. For row actions provide recordSelector. Parameters: expectDialog, waitForDialogMs. Returns: {result: "Success"|"Navigation"|"DialogOpened", navigation?:{pageContextId}, dialog?:{dialogId,fields}}. Common actions: Edit, New, Delete, Post, Save, Cancel, OK. Errors: ActionUnavailable, RecordNotFound, DialogTimeout.
Retrieves metadata for a Business Central page and initializes a page context for subsequent operations. Returns pageContextId (required for stateful operations), page type (List/Card/Document/Worksheet), fields with metadata (id, caption, dataType, editable, required), actions, and page structure. Side effects: Creates a short-lived pageContextId for maintaining state across operations.
No output schemas documented. LLMs cannot predict response structure (e.g., what fields does get_page_metadata return? Is it {pageContextId, fields[], actions[]}?). Descriptions mention return values but no formal schema.
No error handling guidance. Tools document error types (ActionUnavailable, RecordNotFound, DialogTimeout) but do not tell LLMs what to do next. Missing recovery hints like 'If ActionUnavailable, verify the page is in edit mode' or 'If RecordNotFound, try search_records() first'.
Tool annotations missing. execute_action is marked IRREVERSIBLE and create_record_by_field_name is WRITE, but no destructiveHint or idempotentHint in schema. LLMs cannot infer safety properties without explicit annotations.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
execute_action uses oneOf constraint requiring either pageContextId OR pageId. This is correct for mutual exclusivity but LLMs often struggle with oneOf logic. Should add explicit description: 'Provide EITHER pageContextId (preferred) OR pageId (fallback if multiple instances not open).'
create_record_by_field_name 'fields' parameter has additionalProperties type string but no format/pattern constraints. Descriptions mention SourceExpr syntax and scoped names (General > Name) but no formal validation rules. LLMs cannot know valid syntax without examples or patterns.