Real-time trustworthiness evaluation and safety interception for AI agents via MCP
Three tools with basic schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. verify_action has the most complete schema with 5 typed parameters, each with descriptions. get_policy_rules and run_benchmark have empty input schemas ({}), which is appropriate for read-only discovery tools but limits composability. Descriptions are present but generic (10-50 chars), below the 50-200 char baseline for LLM-optimized tools. No error handling guidance, no output schema documentation, and no tool annotations (readOnlyHint present in metadata but not in schema). Parameter descriptions exist but lack format/constraint details (e.g., 'Type of action' does not enumerate valid action_type values). No evidence of pagination, batching, or chaining support.
Return the current policy rules as JSON.
Run the AgentTrust benchmark suite and return a summary as JSON.
Verify an agent action and return a trust report as JSON.
Output schemas not documented. LLMs cannot plan downstream calls or extract required fields without knowing what verify_action, get_policy_rules, and run_benchmark return.
Parameter descriptions lack format/constraint details. 'Type of action' does not enumerate valid action_type values (file_read, shell_command, network_request, etc.). LLMs will hallucinate invalid values.
Tool descriptions are too short (10-50 chars). Below 50-char baseline, they lack context for LLM selection. 'Verify an agent action and return a trust report as JSON' (54 chars) is borderline; 'Return the current policy rules as JSON' (39 chars) is insufficient.
No error handling guidance. If verify_action encounters an invalid action_type or policy evaluation fails, the tool provides no recovery hint. LLMs cannot self-correct.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
No tool annotations in schema. readOnlyHint and idempotentHint are declared in metadata but not in the JSON Schema inputSchema. Clients cannot discover these hints programmatically.