MCP server for managing Azure virtual machines through a Microsoft Teams bot interface, with session management and auto-shutdown capabilities
Server has 5 tools with basic schemas and descriptions, but significant gaps prevent higher scores. Tool names follow verb_noun convention (list_vms, get_vm_status, start_vm, stop_vm, restart_vm) which is good. However, parameter descriptions are minimal or absent in visible schema definitions. The mcp-server/main.py excerpt shows ToolDefinition models registered but parameter descriptions are not visible in the truncated code. Input schemas exist but lack detailed parameter documentation. Error handling is present in the middleware but not evident in tool definitions themselves. No tool annotations (readOnlyHint, destructiveHint) visible despite clear risk levels (READ_ONLY, WRITE, DESTRUCTIVE). Output schemas are not documented.
Get detailed status of a specific VM including power state, network interfaces, and IP addresses.
List all virtual machines in the configured Azure resource group with their power state, size, and OS type.
Restart a running virtual machine.
Start a stopped/deallocated virtual machine.
Stop and deallocate a running virtual machine. This will release the compute resources and stop billing.
Parameter descriptions missing or incomplete. vm_name parameter in get_vm_status, start_vm, stop_vm, restart_vm has minimal description ('The name of the virtual machine'). No guidance on format, constraints, or how to discover valid VM names.
No output schema documentation. Tools return results but LLM cannot know what fields to expect (e.g., does list_vms return vm_id, vm_name, power_state, size, os_type?). This forces LLMs to guess and breaks downstream tool chaining.
No tool annotations despite clear risk levels. stop_vm is DESTRUCTIVE (deallocates resources, stops billing) but has no destructiveHint annotation. start_vm is WRITE but lacks annotation. list_vms and get_vm_status are READ_ONLY but lack readOnlyHint. Annotations guide agent safety reasoning.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | <=2025-11-25 | v2 |
No error recovery guidance in tool descriptions. If a VM name is invalid or not found, the LLM has no hint to call list_vms first. Descriptions should state 'Call list_vms to discover available VM names' for tools that require vm_name.
No confirmation/dry-run pattern for destructive operations. stop_vm deallocates resources and stops billing, a critical action. Should support a dry-run or require explicit confirmation to prevent accidental execution.