28 tools with basic schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. Most tools have 1-2 required parameters with type and description, meeting minimum baseline. However, 14 tools lack any description of what they return, and many descriptions are generic (10-30 chars). No tool annotations (readOnlyHint/destructiveHint) despite clear risk stratification. Parameter descriptions are minimal, e.g., 'Name of the node' repeated 20+ times without context on format or constraints. No pagination support documented for list operations (get_nodes, get_vms, get_containers). Error handling and recovery guidance absent. Composition is reasonable (verb_noun naming, single responsibility), but output chaining IDs missing, e.g., get_vms returns VM list but unclear if vmid is included for downstream start_vm calls.

• Add output schema documentation to every tool. For get_vms, document: returns array of {vmid: integer, name: string, status: string, memory: integer, cores: integer}. This enables LLMs to chain calls (e.g., get_vms → start_vm using vmid).
• Add tool annotations: mark get_* tools with readOnlyHint=true; mark delete_vm, stop_vm with destructiveHint=true; mark create_vm, clone_vm with idempotentHint=false (or true if retries are safe).
• Expand parameter descriptions with format and constraint guidance. Replace 'Name of the node' with 'Node name (alphanumeric, 1-64 chars, e.g., pve-node-1). Obtain from get_nodes().'
• Add error handling guidance to destructive tools. E.g., delete_vm: 'Returns error if VM is locked or running (unless force=true). If force fails, VM may be in inconsistent state, check get_vm_status before retrying.'
• Add pagination to list tools. E.g., get_vms: add limit (1-100, default 20) and offset (default 0) parameters; return {vms: [...], total: integer, has_more: boolean}.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry