Loading ToolBench
Community Model Context Protocol (MCP) server for Cisco Security Cloud Control (Platform, Mesh Policy Engine, Firewall Manager, AI Defense Management).
Two tools with complete schemas and reasonable descriptions. Both are read-only list/get operations with proper pagination support (limit/offset). Naming follows verb_noun convention (aid_list_*, aid_get_*). However, descriptions are brief (under 100 chars) and lack context about when to use each tool or what distinguishes them. Parameter descriptions are minimal. No output schema documentation visible. Error handling and recovery guidance absent. Schemas are present and typed but lack detail on response structure.
Fetch a single AI Defense application by id. Read-only.
List AI Defense registered applications for the tenant. Supports limit/offset pagination and optional expanded objects. Read-only.
Output schemas not documented. LLMs cannot infer what fields are returned or plan downstream tool calls. Both tools lack response structure documentation.
Tool descriptions lack context about when to use each. 'List AI Defense registered applications' vs 'Fetch a single AI Defense application', no guidance on whether to call list first or jump to get_application. Descriptions should explain discovery vs direct access.
Parameter descriptions are minimal. 'expanded' parameter lacks explanation of what 'expanded objects' means, what additional fields are included? When should an LLM request expanded vs compact?
No error handling or recovery guidance. If application_id is invalid or not found, what should the LLM do? Should it call aid_list_applications to discover valid IDs? No guidance provided.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | <=2025-11-25 | v2 |
Pagination limit defaults to 50 but no maximum stated. Can an LLM request limit=10000? Should results be capped? Unbounded pagination can cause context window exhaustion.