Self-hosted remote MCP server for Power BI. Workaround for the broken Microsoft-hosted Power BI MCP endpoint which fails with AADSTS9010010. Performs OAuth proxy via FastMCP's AzureProvider and enables On-Behalf-Of token exchange so Power BI RLS applies per user.
Three tools with clear verb-noun naming (list_*, run_*) and documented input schemas. Descriptions are present but brief (34 - 89 chars, below the 50 - 200 char baseline for LLM optimization). All parameters have type definitions and descriptions. Output schemas are not documented, LLMs cannot predict what fields to expect from list_workspaces, list_datasets, or run_dax_query. Error handling is absent; no guidance on retryability, user-fixable vs. fatal errors, or recovery paths. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Composition is sound: tools chain logically (discover workspaces → datasets → query). Security is strong: no secrets in parameters, OBO token exchange via server-side injection, stateless HTTP enabled.
List datasets in a Power BI workspace. Returns name and ID for each.
List Power BI workspaces you have access to. Returns name and ID for each.
Execute a DAX query against a Power BI dataset. Use list_workspaces and list_datasets first to find workspace_id and dataset_id. The query must start with EVALUATE.
Output schemas not documented. LLMs cannot predict what fields list_workspaces, list_datasets, or run_dax_query return. This forces agents to guess at response structure and risks failed downstream tool calls.
Descriptions are too brief (34 - 89 chars vs. 50 - 200 char baseline). 'List Power BI workspaces you have access to. Returns name and ID for each.' lacks context on when to call this vs. list_datasets, and does not explain the structure of returned items.
No error handling guidance. If run_dax_query fails (invalid DAX, dataset not found, permission denied), the tool returns no recovery hint. LLM cannot distinguish retryable from fatal errors.
No tool annotations. All three tools are read-only (Risk: READ_ONLY declared in metadata), but readOnlyHint is not set in tool definitions. LLMs cannot infer safety from the tool definition alone.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
run_dax_query description includes example constraint ('must start with EVALUATE') but does not explain why or what happens if violated. No validation error message template provided.