Loading ToolBench
Local-only MCP bridge for the aipermission gateway. Provides tools for vault operations and local permission management.
AIPermission MCP defines 3 tools with basic structure but significant gaps in description quality and parameter documentation. Tool names follow verb_noun convention (list_, call_, query patterns). Schemas are present but descriptions are terse (10-80 chars, below the 50-200 char baseline for LLM optimization). Parameters lack depth: 'project_ref' descriptions are minimal; 'input' in call_vault_action is vague ('Action input. Never include raw secret values.') without specifying what structure is expected. No output schemas documented. Error handling and recovery guidance absent. The 'reason' parameter in call_vault_action is good practice (audit trail), but overall parameter documentation is sparse.
Call a vault action (generate_item or restart_session_with_environment) with the specified input
List vault items, optionally filtered by project reference
Query the status or result of a vault action request
Parameter descriptions are too brief and lack actionable constraints. 'project_ref' description ('Optional project reference. Prefer id:<id> or slug:<slug>; omit to list every readable project.') is 95 chars but does not explain format, validation rules, or error cases. 'input' in call_vault_action is vague ('Action input. Never include raw secret values.'), what structure? What fields are required?
No output schemas documented. LLMs cannot plan downstream calls or extract required fields (e.g., does list_vault_items return item IDs? Does call_vault_action return a request_id for polling?). This forces agents to guess or make extra discovery calls.
call_vault_action accepts an 'input' parameter of type 'object' with no schema or structure guidance. Different actions (generate_item vs restart_session_with_environment) likely require different fields, but this dependency is undocumented. LLMs cannot validate input without explicit schema.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
No error handling or recovery guidance. If call_vault_action fails (invalid action, missing required fields, permission denied), the tool provides no actionable error message or next steps. Agents cannot self-correct.
Tool descriptions are below the 50-200 char baseline for LLM optimization. 'List vault items, optionally filtered by project reference' (60 chars) lacks context on when to call this vs other discovery tools, what structure is returned, and whether pagination is supported.