HTTP-based MCP server providing tools to query and manage Microsoft Dataverse/Power Platform entities, attributes, relationships, option sets, and records
The Dataverse MCP server has 8 tools with complete schema definitions and descriptions, but the quality is mixed. All tools are READ_ONLY with basic Zod schemas (z.string(), z.enum(), z.number().optional()). Descriptions are present but often generic and lack guidance on when/why to use each tool. Parameter descriptions are minimal, typically just 'Logical name of the entity' without examples, format constraints, or disambiguation. Tool naming is mostly action-verb-based (get_*, query_*), but several tools could be consolidated. Output handling returns raw JSON.stringify() responses without documented structure, pagination support, or field guidance for chaining. Error handling is minimal, the HTTP endpoint catches errors and returns a 500 with err.message, offering no recovery guidance. No validation of input parameters (e.g., entityName, filter strings). Tool composition has gaps: query-records lacks result limits and pagination; get-record and query-records accept plural names ('accounts', 'contacts') while other tools use singular names, creating inconsistency. No tool annotations (readOnlyHint, idempotentHint) despite all being read-only. No security measures visible (credential injection appears to rely on environment variables, but no explicit validation or scoping). This is a functional but minimally-polished server landing in the D-grade range.
Get a specific attribute from an entity
Get all attributes for a Dataverse entity
Get metadata for a Dataverse entity
Get all relationships for a Dataverse entity
Get a global option set by name
Get a record by ID
Query Dataverse records with OData filter
Output schemas are not documented. All tools return JSON.stringify() with no declared structure. LLMs cannot infer response fields or plan downstream tool calls.
Inconsistent parameter naming: get-record and query-records use 'entityNamePlural' (e.g., 'accounts'), while all other tools use singular 'entityName' (e.g., 'account'). This will cause LLMs to provide wrong parameter types.
query-records accepts an unvalidated OData filter string with no syntax examples or constraints. LLMs will hallucinate invalid OData expressions. Missing pagination/result limit documentation.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 54 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 57 | - | v1 |
Use a predefined prompt for entity overview
Parameter descriptions are minimal or absent. 'Logical name of the entity' appears 5+ times with no clarification of what 'logical name' means to an end user or examples (e.g., 'account' vs 'Account' vs 'accounts').
No error recovery guidance. The HTTP /invoke endpoint catches errors and returns 500 with err.message. No indication of whether an error is retryable, user-fixable, or fatal. No recovery actions suggested.
No tool annotations. All 8 tools are read-only (Risk: READ_ONLY declared), but readOnlyHint is not used in Zod schema definitions.
use-powerplatform-prompt has a hard-coded enum with one value ('ENTITY_OVERVIEW'). This severely limits extensibility and suggests poor separation of concerns.
No input validation before calling PowerPlatformService. entityName, attributeName, filter, and optionSetName are passed as raw strings. Risk of SQL injection or malformed queries if the backend doesn't validate.